FOCI: Trajectory Optimization on Gaussian Splats

• Project page: https://rffr.leggedrobotics.com/works/foci/

Authors

• Mario Gomez Andreu*¹
• Maximum Wilder-Smith*¹
• Victor Klemm¹
• Vaishakh Patil¹
• Jesus Tordesillas²
• Marco Hutter¹

*Equal contribution
¹ Robotic Systems Lab, ETH Zurich
² Comillas Pontifical University

Overview

FOCI is a novel method to compute orientation aware trajectories for robots using 3D Gaussian Splats to model both the robot and the environment.

Installation

0. Install git lfs
1. Clone the repo with git clone [email protected]:leggedrobotics/foci.git
2. Obtain a licence for the MA27 solver from https://licences.stfc.ac.uk/product/coin-hsl, download the corresponding ZIP file, rename the extracted folder to coinhsl and move it to the root directory of this repository.
3. docker build -t rsl/foci . to build the provided docker container.
4. docker run -it -v .:/workspace --gpus all -p 127.0.0.1:8080:8080 rsl/foci to run and attach to the container.
5. pip install -e . to install the foci in the docker container.
6. python3 demos/stonehenge.py to run the demo script. Open 127.0.0.1:8080 in your webbrowser to a see a visualisation similar to the one in this README.md

Citing

If you find this work useful, please consider citing our paper:

@article{andreuwildersmith2025foci,
        author        = {Mario Gomez Andreu and Maximum Wilder-Smith and Victor Klemm and Vaishakh Patil and Jesus Tordesillas and Marco Hutter},
        title         = {FOCI: Trajectory Optimization on Gaussian Splats},
        year          = {2025},
        eprint        = {2505.08510},
        archivePrefix = {arXiv},
        primaryClass  = {cs.RO},
        url           = {https://arxiv.org/abs/2505.08510}
}

Data Attribution

The Gaussian splat file (demo/data/stonehenge.ply) included in this repository was generated using processed data from the Splat-Nav project, which is licensed under the MIT License. We acknowledge the authors of Splat-Nav for making their resources available.