ThiesOelerich/SafeFlowMPC_pretrain

SafeFlowMPC Pretrain Dataset

This repository contains the pretraining dataset for the paper SafeFlowMPC: Predictive and Safe Trajectory Planning for Robot Manipulators with Learning-based Policies.

Project Page | Github Code

Dataset Description

The dataset consists of robotic trajectories used to train a flow matching model for safe trajectory planning. The data was generated using a modified version of the VP-STO planner and is designed for 7-DoF manipulators (specifically tested on KUKA robots).

This specific dataset is used for the pretraining stage of the imitation learning model, which focuses on learning policies without explicit safety considerations. For the finetuning stage with safety considerations, see the SafeFlowMPC dataset.

It contains:

• trajectories: Sampled robotic joint trajectories.
• conditional_data: Contextual information for the planner (e.g., initial configurations and goal poses).

Usage

This dataset is automatically loaded by the training scripts provided in the official repository. To pretrain the model using this dataset, you can run the following command from the repository:

python train_imitation_learning.py

Citation

If you use this dataset in your research, please cite:

@inproceedings{oelerich2026safeflowmpc,
  title={SafeFlowMPC: Predictive and Safe Trajectory Planning for Robot Manipulators with Learning-based Policies},
  author={Oelerich, Thies and Ebmer, Gerald and Hartl-Nesic, Christian and Kugi, Andreas},
  booktitle={IEEE International Conference on Robotics and Automation (ICRA)},
  year={2026}
}