cuRobo

CUDA Accelerated Robot Library

Documentation | Paper

[!NOTE]

cuRoboV2 is a significant rewrite and the public API has changed from cuRobo v1.

If you depend on the v1 API, pin to the v0.7.8 tag.

cuRobo is a CUDA-accelerated library for robot motion generation, built on PyTorch, CUDA, and Warp. It provides GPU-parallel algorithms for forward/inverse kinematics, collision checking, trajectory optimization, geometric planning, GPU-native perception, and whole-body motion generation, scaling from single-arm manipulators to high-DoF humanoids.

Key capabilities:

• Dynamics-aware trajectory optimization with B-spline representation enforcing smoothness and torque limits
• GPU-native ESDF perception that generates dense signed distance fields from depth images, up to 10x faster than state-of-the-art
• Scalable whole-body computation including topology-aware kinematics, differentiable inverse dynamics, and map-reduce self-collision for high-DoF robots
• Collision-free motion generation combining IK, geometric planning, and trajectory optimization.

C++ Runtime

Start with AGENTS.md for the Python-first integration workflow: validate the robot and motion quality, allocate collision caches, warm up once, and reuse solvers.

The cpp_runtime/ directory contains backend-neutral C++ APIs for robot kinematics, goalset inverse kinematics, paired-candidate time-optimal B-spline seed fitting, joint-space trajectory optimization with timestep refinement, paired-lane bidirectional RRT over IK solutions, composed joint-space motion planning, single-pose IK-to-RRT-to-TrajOpt planning, goalset grasp approach/insertion/lift, single-scene collision data, voxel sphere fitting, grasped-object attachment, and the block-sparse mapper. Python remains the source of truth for production cuRobo and Warp kernels, while C++-specific export glue lives under cpp_runtime/robot/kernels/. Generated architecture-specific artifacts are not committed.

Citation

If you found this work useful, please cite cuRoboV2-Fast,

@misc{curobo_v2_fast,
      title={cuRoboV2-Fast: Fast Torque-Limited Collision-Free Motion Planning},
      author={Balakumar Sundaralingam and Adithyavairavan Murali and Stan Birchfield},
      year={2026},
      eprint={2611.06913},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2611.06913}
}

Contributing

Contributions are welcome. Bugs: open an issue. General usage questions: GitHub Discussions. For pull requests, please read CONTRIBUTING.md. All commits must include a DCO sign-off (git commit -s).

License

cuRobo is released under the Apache 2.0 license.

The example robot assets bundled in this repository are provided under their respective licenses. See LICENSE_ASSETS for details.

| Asset directory | License | Local license file | | --- | --- | --- | | franka_description | Apache-2.0 | LICENSE | | g1 | BSD-3-Clause | LICENSE | | ur_description | BSD-3-Clause | LICENSE |

Third-Party Software

This project will download and install additional third-party open source software projects. Review the license terms of these open source projects before use.

Dependency licenses and attributions are listed in THIRD_PARTY_NOTICES.md.

NOTICE AND DISCLAIMER: This software automatically retrieves, accesses or interacts with external materials. Those retrieved materials are not distributed with this software and are governed solely by separate terms, conditions and licenses. You are solely responsible for finding, reviewing and complying with all applicable terms, conditions, and licenses, and for verifying the security, integrity and suitability of any retrieved materials for your specific use case. This software is provided "AS IS", without warranty of any kind. The author makes no representations or warranties regarding any retrieved materials, and assumes no liability for any losses, damages, liabilities or legal consequences from your use or inability to use this software or any retrieved materials. Use this software and the retrieved materials at your own risk.