Reactive Collision-Aware Motion Planning for 7-DOF Manipulator Using NVIDIA cuRobo and Live Point Cloud Perception · Autonomous Mobile Robotics Laboratory, UT Austin

Developing control pipeline to translate natural language high-level goals into executable tasks for a 7-DOF robotic arm
Implementing GPU-accelerated, collision-free motion planning via the MotionGen trajectory optimizer and a reactive MPC controller in cuRobo v2 against an ESDF environment fused from 3D LiDAR and depth camera point clouds
Preparing deployment of the autonomous system on NVIDIA Jetson AGX Thor with validated 90 ms global trajectory generation latency, 25 Hz worst case MPC control loop frequency, and pending test for collision avoidance accuracy