Anytime GTMP

Sai Coumar. An T. Le, Zachary Kingston

Anytime GTMP extends Global Tensor Motion Planning (GTMP) with connectorized graph edges generated by local planners from VAMP, enabling robust connections between sampled states. It formulates planning as an anytime, asymptotically optimal (AO) problem under a fixed time budget, progressively improving solution quality as computation time increases.

Setup:

Download Anytime Gtmp

git clone https://github.com/commalab/anytime_gtmp
cd anytime_gtmp
git submodule update --init --recursive
cd vamp
git checkout benchmark_aorrtc_backend
cd ..

Install Dependencies Anytime Gtmp

uv venv --python 3.11
source .venv/bin/activate
uv pip install -e ./pyroffi
uv pip install -e ./vamp
uv pip install -r requirements.txt

Citation

If you found this work useful, please consider citing:

@misc{coumar2026anytimeglobaltensormotion,
      title={Anytime Global Tensor Motion Planning}, 
      author={Sai Coumar and An T. Le and Zachary Kingston},
      year={2026},
      eprint={2608.25830},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2608.25830}, 
}

License

Our code is open source under the MIT license. See LICENSE for details.