cuTAMP: Differentiable GPU-Parallelized Task and Motion Planning
-
Updated
May 30, 2026 - Python
cuTAMP: Differentiable GPU-Parallelized Task and Motion Planning
Example manipulation behaviors using behavior trees, as well as Docker workflows in ROS2 based projects.
Mixed Palletizing with Explainable Visual Reasoning
GPU-accelerated robot motion planning on Apple Silicon. Port of NVIDIA cuRobo (CUDA) to MLX — real-time collision-free trajectory generation on M-series Macs.
Dexterous pick-and-place with Allegro hands on a dual-arm UR3/UR3e: FoundationPose 6D perception + BODex grasp synthesis, in Isaac Lab and on real hardware
ROS 2 active-view 3D scanning with a PiPER arm, RealSense L515, GroundingDINO, SAM 2, NBV planning, Tesseract/cuRobo, and fail-closed execution.
Development of an end-to-end robotic manipulation pipeline using ROS 2, Isaac Sim, and the Unitree G1 humanoid. The system performs semantic object detection, grasp prediction, collision-aware motion planning, and autonomous pick-and-place for supermarket shelf automation.
Camera-guided pick and place for a UR5 + Robotiq 2F-85 in Isaac Sim 5.1, planned by cuRobo 0.8 — the planner is never told where the obstacle is. Includes a gymnasium env for RL.
Web jog panel, teach mode and programs for the Umbratek UTRA 6-1000 robot arm, with a simulator and MuJoCo twin to try it without the arm
To associate your repository with the curobo topic, visit your repo's landing page and select "manage topics."