Skip to content
leggedroboticsPublic

About

Fast, orientation-aware trajectory planning using a novel Gaussian overlap-based collision formulation, modeling both robot and environment as Gaussian Splat.

Topics

Resources

Stars

56 stars

Watchers

4 watching

Forks

Latest commit

 

History

92 Commits

Folders and files

Repository files navigation

FOCI: Trajectory Optimization on Gaussian Splats for Gaussian Robots

Static Environment
Stonehenge paths
Dynamic Environment
Rock and barrels course

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

Authors

  • Mario Gomez Andreu*1
  • Carlos Martínez*2
  • Maximum Wilder-Smith*1
  • Victor Klemm1
  • Vaishakh Patil1
  • Jesus Tordesillas2
  • Marco Hutter1

*Equal contribution
1 Robotic Systems Lab, ETH Zurich
2 Trustworthy Robotics and AI Lab (TRAIL), ICAI-Comillas

Citation

When using FOCI, please cite:

@inproceedings{andreu2025foci,
  title={Foci: Trajectory optimization on gaussian splats},
  author={Andreu, Mario Gomez and Klemm, Victor and Patil, Vaishakh and Tordesillas, Jesus and Hutter, Marco},
  booktitle={2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={1884--1891},
  year={2025},
  organization={IEEE}
}

Installation

  1. Clone the repo with git clone git@github.com: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. Create the virtual environment with python3 -m venv venv and install dependencies with pip install -r requirements.txt.
  4. Download the Gaussian splat scenes listed under Data Attribution and place them in the data/ directory (these .ply files are not tracked by git, see .gitignore).
  5. Launch any demo file with python3 -m demos.<demo_name>.

Data Attribution

The Gaussian splat file (data/stonehenge.ply) used by 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.

The following scenes were obtained from the Polycam community gallery and are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0):

File Original capture
data/Barrel.ply Barrel
data/Rock on forest path.ply Rock on forest path
data/Bonsai.ply new bonsai
data/Coffee_Table.ply Coffee Table
data/Forest.ply Forest Scan

These scenes are third-party assets and are not covered by the license of this repository's source code. We thank the Polycam community members for sharing their captures.

About

Fast, orientation-aware trajectory planning using a novel Gaussian overlap-based collision formulation, modeling both robot and environment as Gaussian Splat.

Topics

Resources

Stars

56 stars

Watchers

4 watching

Forks

Releases

Packages

Contributors

Languages