Open Source RoboticsRobots, software & learning

cuRobo

Software

CUDA-accelerated motion planning library for robot kinematics, collision checking and trajectory optimization; the reviewed v0.8.0 release introduces the V2 API. [1]

Installation & first result →Official websiteGetting started

For
Robotics developers integrating GPU motion planning into a configured robot stack [1] [3]
Hardware
Franka Panda model in the motion planning tutorial [3]
Source rights
Named licenses are documented by part. Review licenses
Unknowns
compute requirements not fully established. Review gaps

Source researched · Not tested by us
Sources checked · Research details

Licenses & related hardware

Requirements · Research details and dates

On this page

Overview

cuRobo plans robot motion on an NVIDIA GPU: it combines inverse kinematics, collision checking and trajectory optimization to turn a joint start state and a target pose into a candidate trajectory. The documented Panda example is a practical first result because it reports planning success and saves trajectory plots without requiring a physical robot connection. [1] [3]

Is this for you?

Known limitations
  • V2 changes most existing APIs; v1 examples require their matching version.
  • Robot assets have separate licenses.
  • A planned trajectory does not establish hardware controller integration or physical safety.

Editorial assessment from cited documentation; no runtime testing.

[1] [3] [4] · Editorial assessment

What you need

Operating system
  • Ubuntu >=20.04; installation walkthrough uses Ubuntu 22.04

[2]

Languages / runtimes
  • Python
  • CUDA

[1] [5]

Dependencies
  • Python >=3.10; >3.13 not validated in installation guide
  • PyTorch via CUDA-specific extras
  • Warp
  • CUDA 12 or 13 runtime extras

[2] [5]

Host computer

Upstream specifies an NVIDIA GPU newer than Turing, at least 4 GB VRAM and driver >=580.65.06. This is not a workload performance guarantee.

[2]

Installation & first result

Open the upstream getting-started guide [2] · Project documentation [1]

  • Check out v0.8.0 and create a Python 3.11 uv environment.
  • Select the CUDA 12 or CUDA 13 extra, with PyTorch included for a fresh environment.
  • Run the import check and upstream pytest suite.
  • Run the motion_planning example and inspect success flags and saved trajectory PDFs.

[2] [3]

Installation

Use the documented uv installation with matching cu12-torch or cu13-torch extras for a fresh environment.

[2]

Hardware, models & data

Supported hardware
  • Franka Panda model in the motion planning tutorial; physical controller compatibility not verified

[3]

Inputs and outputs

Joint start states, target poses and obstacles produce planning success indicators and joint trajectories; tutorial saves PDF plots.

[3]

Evaluation

Upstream import and pytest checks plus a Panda planning tutorial are documented; this catalog has not run them.

[2] [3]

Project notes

Match the API before following an example

This profile reviews release v0.8.0, published April 18, 2026, at commit 4ea77366ca48ee453e7df139e39fa6532af49f3b. Its release notes describe a major refactor that breaks most existing APIs. The README directs users dependent on the older v1 API to tag v0.7.8. Keep tutorials, configuration and code on a compatible revision; do not mix a V2 install with an older integration just because both are called cuRobo. [6] [1]

Install a matching GPU environment

The release's installation guide uses Ubuntu 22.04 and specifies Ubuntu 20.04 or newer, an NVIDIA GPU newer than Turing with at least 4 GB VRAM, and driver 580.65.06 or newer. It requires Python 3.10 or newer, notes that versions above 3.13 are unvalidated, and uses Python 3.11 in its uv example. Those are upstream requirements, not measured performance on a particular robot or scene. [2]

Choose the CUDA 12 or CUDA 13 installation extra to match the documented driver setup. The cu12-torch and cu13-torch extras include PyTorch for a fresh environment; the corresponding extras without -torch assume it is already installed. Follow the import check and upstream pytest command before trying the planning tutorial. [2] [5]

Run python -m curobo.examples.getting_started.motion_planning to exercise pose planning and approach–grasp–lift planning. The tutorial uses a Franka Panda configuration and an obstacle scene, returns success information and saves PDF trajectory plots under the cuRobo cache. Its interactive option displays the example in Viser. These results describe the planning model; they do not verify calibration, controller integration or execution on a real arm. [3]

What we checked

Source researched only; no installation, training, benchmark or physical hardware test performed. No affiliate links or reused media.

Updates

Upstream last updated
Apr 18, 2026 [6]
Latest release
Apr 18, 2026 [6]
Profile updated
Oct 17, 2026
Sources last checked
Oct 17, 2026
Next review
Nov 17, 2026
Reviewed revision
v0.8.0 / 4ea77366ca48ee453e7df139e39fa6532af49f3b · software repository

Source files and licenses

Open motion planning code with separately licensed robot-description assets; physical integration is not established by catalog inclusion.

LayerAvailability / license
softwareavailable · Apache-2.0
Repository materials only; external components retain their own terms. [7] [4]
documentationavailable · Apache-2.0
Repository materials only; external components retain their own terms. [7]

The reviewed planning-library workflow is not a trained-model distribution.

The reviewed workflow uses robot descriptions and planning scenes, not a training dataset; asset terms are recorded separately.

License scope and remaining work

Project code is Apache-2.0. Bundled robot descriptions retain separate terms: LICENSE_ASSETS records Franka material under Apache-2.0 and Universal Robots and Unitree G1 material under BSD-3-Clause. Check the relevant asset terms when reusing a robot model. [7] [4]

We reviewed source and documentation, not a running GPU installation. Planning time, success rate for your scene, GPU operating cost and our ability to support a physical integration remain unmeasured. Start with the included model before introducing a custom robot and collision geometry.

References

Each source links to the exact revision we read. “Used for” shows where this page relies on it.

Correction guidance · Discussion is not enabled · Profile format 0.7.0-draft.5

Change history