prompt
list | completion
list | images
images list |
|---|---|---|
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "Instruction",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "Instruction",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "Confirm",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOther",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <pickup Knife>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOnObjectDetails",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <pickup Knife>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: there are 3 loaves of bread <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "Instruction",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <pickup Knife>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: there are 3 loaves of bread <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: choose any and slice 1 please <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <pickup Knife>\n",
"type": "text"
},
{
"text": "COMMANDER action: there are 3 loaves of bread <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: choose any and slice 1 please <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <slice Bread>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "Instruction",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <pickup Knife>\n",
"type": "text"
},
{
"text": "COMMANDER action: there are 3 loaves of bread <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: choose any and slice 1 please <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <slice Bread>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: then take a slice of bread <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "Instruction",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <pickup Knife>\n",
"type": "text"
},
{
"text": "COMMANDER action: there are 3 loaves of bread <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: choose any and slice 1 please <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <slice Bread>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: then take a slice of bread <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: and toast it <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "InformationOther",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <pickup Knife>\n",
"type": "text"
},
{
"text": "COMMANDER action: there are 3 loaves of bread <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: choose any and slice 1 please <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <slice Bread>\n",
"type": "text"
},
{
"text": "COMMANDER action: then take a slice of bread <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: and toast it <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: you will need to put the knife down first <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "OBSERVE",
"type": "text"
}
],
"role": "assistant"
}
] | |
[
{
"content": [
{
"text": "## Scenario\nYou are a commander, instructing a household robot to perform various tasks. The robot, named \"follower\" or \"driver\",\nis designed to assist you with household chores and tasks. You will communicate with this robot in a conversational manner,\nanswer its questions, and provide instructions/corrections as needed. The robot will respond to your commands.\n\nYou will be presented with a goal, such as \"make a sandwich\", and a dialogue history between you (the commander) and the\nrobot (the follower/driver). This dialogue history will include actions, instructions, and responses from both parties.\nWhen either you or the robot is speaking, the other party will be silent, displayed as \"OBSERVE\". Otherwise, the action\nwill be displayed. When the action is a dialogue action, the utterance will be displayed, followed by a tag(s) describing\nthe purpose/meaning of the action. This tag is called a dialogue act.\n\n## Dialogue Acts\nEach dialogue action in the history will be labeled with a dialogue act describing the nature of the interaction. They\nare described below. They are given a category, an example, and a description of which party (commander or follower)\ntypically uses them.\n\nThe possible dialogue acts are:\n\n```py\ndialogue_acts = [\n 'Acknowledge', # Category: Generic. Example: \"perfect.\" Mostly follower\n 'Affirm', # Category: Generic. Example: \"yes.\" Mostly commander\n 'AlternateQuestions', # Category: Object/Location. Example: \"yellow or blue mug?\" Mostly follower\n 'Confirm', # Category: Generic. Example: \"Should I clean the cup?\" Mostly follower\n 'Deny', # Category: Generic. Example: \"no.\" Mostly commander\n 'FeedbackNegative', # Category: feedback. Example: \"that is not correct.\" Almost entirely commander\n 'FeedbackPositive', # Category: feedback. Example: \"great job.\" Almost entirely commander\n 'Greetings/Salutations', # Category: Generic. Example: \"hello.\" Even commander/follower\n 'InformationOnObjectDetails', # Category: Object/Location. Example: \"knife is behind the sink.\" Almost entirely commander\n 'InformationOther', # Category: Generic. Example: \"Mug is already clean.\" Mostly commander\n 'Instruction', # Category: Instruction. Example: \"fill the mug with coffee.\" Almost entirely commander\n 'MiscOther', # Category: Generic. Example: \"ta-da.\" Even commander/follower\n 'NotifyFailure', # Category: Interface. Example. \"not able to do that.\" Almost entirely follower\n 'OtherInterfaceComment', # Category: Interface. Example: \"which button opens drawer?\" Even commander/follower\n 'RequestForInstruction', # Category: Instruction. Example: \"what should I do today?\" Almost entirely follower\n 'RequestForObjectLocationAndOtherDetails', # Category: Object/Location. Example: \"where is the mug?\" Almost entirely follower\n 'RequestMore', # Category: Instruction. Example: \"Is there anything else to do?\" Almost entirely follower\n 'RequestOtherInfo', # Category: Instruction. Example: \"how many slices of tomato?\" Almost entirely follower\n]\n``` \n\n## Visual information\nYou may be given visual information about your viewpoint, the robot's viewpoint, and/or the target object's viewpoint.\nUse this information to decide whether to actively correct the robot's actions, to observe the robot's actions, or to\ninteract with the robot as you see fit.\n\n## Your Task\nGiven this dialogue history, your task is to predict the dialogue act for the next turn in the dialogue in order\nto accomplish the given goal.",
"type": "text"
}
],
"role": "system"
},
{
"content": [
{
"text": "GOAL: Make a plate of toast.\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: directions please <<RequestForInstruction>>\n",
"type": "text"
},
{
"text": "COMMANDER action: today we need to make a plate of toast <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the bread is on the counter <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: by the lettuce <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: there should be a knife <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: in the drawer under the toaster <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: the toaster is by the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: open the drawer under the toaster <<Instruction,InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: the knife should be in there <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: do you see it? <<Confirm>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <close Drawer>\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: no knife <<Deny>>\n",
"type": "text"
},
{
"text": "COMMANDER action: there should also be a knife in the fridge <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: lets check there <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <open Fridge>\n",
"type": "text"
},
{
"text": "COMMANDER action: in the freezer <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <pickup Knife>\n",
"type": "text"
},
{
"text": "COMMANDER action: there are 3 loaves of bread <<InformationOnObjectDetails>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: choose any and slice 1 please <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: <slice Bread>\n",
"type": "text"
},
{
"text": "COMMANDER action: then take a slice of bread <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER action: and toast it <<Instruction>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: you will need to put the knife down first <<InformationOther>>\n",
"type": "text"
},
{
"text": "DRIVER action: OBSERVE\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER action: OBSERVE\n",
"type": "text"
},
{
"text": "DRIVER action: having trouble with knife <<NotifyFailure>>\n",
"type": "text"
},
{
"text": "COMMANDER viewpoint:\n",
"type": "text"
},
{
"text": null,
"type": "image"
},
{
"text": "\n",
"type": "text"
},
{
"text": "COMMANDER's Response: ",
"type": "text"
}
],
"role": "user"
}
] |
[
{
"content": [
{
"text": "Instruction",
"type": "text"
}
],
"role": "assistant"
}
] |
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 47