GPU robotics toolkit

CudaRobotics Docs

Choose a browser, Python, or Nav2 path and produce a checked result before moving into the full benchmark and research surface. The browser package build and release assets remain bound to published v0.3.0.

No local GPU: run Colab Python on my GPU ROS 2 / Nav2

Fast Paths

Choose the entry point that matches how you want to try the project.

Browser

Colab quickstart

Run the latest onboarding cells in 5–10 minutes; the package build remains pinned to published v0.3.0.

Open latest Colab onboarding

Python

One-command local check

Run MPPI and registration together and retain a success or classified failure result.

Start the Python path

ROS 2

Nav2 and CudaNav

Move from the controller plugin to a deterministic closed-loop smoke with a retained manifest.

Open the Nav2 path

Core Guides

Python MPPI and Registration

`MppiPlanner`, CUDA DLPack costmaps, FilterReg, robust registration, Sinkhorn, FGR, and BCPD examples.

Nav2 CUDA MPPI Controller

Plugin setup, parameter validation, motion models, smoke tests, and the controller architecture boundary.

Benchmark Results

Fixed-seed MPPI suites, Nav2 CPU-vs-GPU comparisons, and registration external baselines.

Research Notes

MPPI reproduction zoo, planning demos, perception demos, and reproducibility notes.

Nav2 CUDA MPPI Gazebo demo — Nav2 CUDA MPPI controller running through a Gazebo mission.

v0.3.0 Release

v0.3.0 adds the integrated CudaNav stack, release-bound evidence, reproducible registration suites, and typed registration results, and an evidence-labelled real-rosbag evaluation workflow.

Area	What is included	Where to go
Python	GPU MPPI plus typed and tuple-compatible point-cloud registration APIs.	Python API
Nav2	CUDA MPPI controller plugin with DiffDrive, Ackermann, and Omni support.	Nav2 guide
Distribution	GitHub release assets, Colab quickstart, and GHCR demo image.	Install guide
Evidence	Quality-gated registration and recorded/shadow rosbag reports with visible failures.	Results index