MotoVLA: Generalist Robot Manipulation Beyond Action Labeled Data

[Project] [Paper] [Models]

News

[2025.09.15] We release model weights [https://huggingface.co/aspiridonov/motovla/tree/main], evaluation and inference code.

[2025.08.04] MotoVLA was accepted in CoRL2025!

MotoVLA

Getting Started

# Clone Repo
git clone --recurse-submodules [email protected]:insait-institute/motovla.git
cd motovla

# Create Conda Env
conda create -n eval python=3.10
conda activate eval

# Install Jax
pip install --upgrade "jax[cuda]==0.6.2" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

# Install SimplerEnv
pip install -e third_party/SimplerEnv/ManiSkill2_real2sim/
pip install -e third_party/SimplerEnv/

# Install other dependecies
pip install -r requirements.txt

Next, download the model checkpoint from Huggingface repository. Download, the files under checkpoints directory.

python scripts/download_checkpoint.py

Running Inference in SimplerEnv

Running a single task in simpler:

python -m scripts.eval --config src/configs/motovla/eval_config.py:task="widowx_carrot_on_plate"

Running full simpler evaluation:

sbatch slurm/eval_simpler_bridge.sh

Running Inference on WidowX robot

Run this command and query the model from within the same local network.

python -m scripts.hosting_motovla

Citation

@inproceedings{
  spiridonov2025generalist,
  title={Generalist Robot Manipulation beyond Action Labeled Data},
  author={Alexander Spiridonov and Jan-Nico Zaech and Nikolay Nikolov and Luc Van Gool and Danda Pani Paudel},
  booktitle={9th Annual Conference on Robot Learning},
  year={2025},
  url={https://openreview.net/forum?id=ZqBXnR6ppz}
}

License

This project is licensed under the MIT License. Some source files are derived from code originally licensed under the Apache License, Version 2.0. See LICENSE-APACHE for details.