multitask-dit-robosuite-wrist-diffusion-018000

LeRobot Multi-task DiT policy trained on robosuite Lift, PickPlaceCan, and NutAssemblySquare with agentview and wrist camera inputs.

Source checkpoint: /home/anthonyliang/Documents/rfm_rl/outputs/multitask_dit_robosuite_wrist_diffusion_gpu0/checkpoints/018000/pretrained_model

Rollout Eval

Overall success rate: 0.7777777777777778 over 90 episodes.

task	episodes	success_rate	avg_return	avg_steps
can	30	0.8333333333333334	24.04904609928033	159.1
lift	30	0.9	14.830980708763446	86.1
square	30	0.6	53.264552917941145	261.1666666666667

Loading

Use this as a standard LeRobot pretrained policy checkpoint.

from lerobot.configs import PreTrainedConfig
from lerobot.policies import get_policy_class

cfg = PreTrainedConfig.from_pretrained("TRI-ML/multitask-dit-robosuite-wrist-diffusion-018000")
policy = get_policy_class(cfg.type).from_pretrained("{repo_id}", config=cfg)