---
license: mit
pipeline_tag: robotics
tags:
- safety
- constraints
- imitation_learning
- flow_matching
- robotics
---

# SafeFlowMPC: Predictive and Safe Trajectory Planning for Robot Manipulators with Learning-based Policies

SafeFlowMPC is a framework that combines flow matching and online optimization to ensure safety for robot manipulators using learning-based policies. This method provides rigorous safety guarantees at all times and is designed for real-time execution.

- **Paper:** [SafeFlowMPC: Predictive and Safe Trajectory Planning for Robot Manipulators with Learning-based Policies](https://huggingface.co/papers/2602.12794)
- **Project Page:** [ACIN - SafeFlowMPC](https://www.acin.tuwien.ac.at/en/42d6)
- **GitHub Repository:** [TU-Wien-ACIN-CDS/SafeFlowMPC](https://github.com/TU-Wien-ACIN-CDS/SafeFlowMPC)

## Usage

To use these models, please refer to the official [GitHub repository](https://github.com/TU-Wien-ACIN-CDS/SafeFlowMPC).

### Installation

```bash
pip install -r requirements.txt
pip install -e .
```

### Running Inference

You can run the example experiments using the provided global planner script. The script is configured to automatically load the required checkpoints from Hugging Face if they are not found locally.

```bash
python inference_global_planner.py
```

Optionally, you can enable random replanning with the `--replan` flag:

```bash
python inference_global_planner.py --replan
```

## Citation

```bibtex
@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}
}
```