---
license: mit
task_categories:
- robotics
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: trajectories
    list: float64
  - name: c_data
    list: float64
  - name: samples
    list:
      list:
        list:
          list: float64
  - name: dsamples
    list:
      list:
        list:
          list: float64
  - name: t_samples
    list:
      list: float64
  splits:
  - name: train
    num_bytes: 187797344364
    num_examples: 199099
  download_size: 177769694322
  dataset_size: 187797344364
---

# SafeFlowMPC Finetuning Dataset

This repository contains the finetuning dataset used in the paper [SafeFlowMPC: Predictive and Safe Trajectory Planning for Robot Manipulators with Learning-based Policies](https://huggingface.co/papers/2602.12794). 

SafeFlowMPC is a method that combines flow matching and online optimization to achieve safe and flexible robotic manipulation. This specific dataset is used for the finetuning stage to incorporate safety considerations into the model.

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

## Usage

This dataset is designed to be used with the SafeFlowMPC training scripts. To finetune a model on this dataset with safety considerations, you can use the following command from the official repository:

```bash
python train_imitation_learning_safe.py
```

The script is configured to load this dataset automatically from the Hugging Face Hub.

## Dataset Creation
The dataset was created using the VP-STO planner. It contains safe intermediate trajectories generated to train the flow matching model for precise and safe robotic tasks, such as grasping and human-robot object handovers on a KUKA 7-DoF manipulator.