GATO

GPU-Accelerated Trajectory Optimization

Numerical experiments and the open-source solver from "GATO: GPU-Accelerated and Batched Trajectory Optimization for Scalable Edge Model Predictive Control"

Installation

git clone https://github.com/A2R-Lab/GATO.git
cd GATO

Docker is used for containerization and uv is used as a Python package/project manager.

Setup

./tools/install.sh

Docker

./tools/docker.sh
'''

Manual Installation
'''sh
git submodule update --init --recursive
uv sync
source .venv/bin/activate
docker build -t gato . # build image
docker run -d -it --gpus all --network host -e DISPLAY=:0 -v $(pwd):/workspace -v /tmp/.X11-unix:/tmp/.X11-unix --name gato-container gato # run container
docker exec -it gato-container bash # enter container
docker exec -it --workdir /workspace gato-container bash # enter container in the workspace directory

docker stop gato-container && docker rm gato-container # stop and remove

GATO

./tools/build.sh

Build Options

You can control which Python extension modules are built by selecting plant models and horizon lengths at CMake configure time:

mkdir -p build && cd build
cmake -DPLANT="indy7;iiwa14" -DKNOTS="8;32;128" ..
cmake --build . --parallel

• PLANT: semicolon-separated list of plant targets (indy7, iiwa14).
• KNOTS: semicolon-separated list of horizon lengths.

Built Python modules are written to python/bsqp/ as bsqpN{N}_{plant}.so.

Requirements

• Ubuntu 22.04
• CUDA 12.6
• C++17
• gcc 11.4.0
• Python 3.10.12
• Docker 28.1.0

Usage

See batch_sqp.cu for a minimal example of a batched trajectory optimization solve in C++/CUDA. Example Jupyter notebooks using GATO for MPC are in examples/

Related

• The open-source MPCGPU solver
• GRiD, a GPU-accelerated library for computing rigid body dynamics with analytical gradients

Cite

@misc{du2025gatogpuacceleratedbatchedtrajectory,
      title={GATO: GPU-Accelerated and Batched Trajectory Optimization for Scalable Edge Model Predictive Control}, 
      author={Alexander Du and Emre Adabag and Gabriel Bravo and Brian Plancher},
      year={2025},
      eprint={2510.07625},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2510.07625}, 
}