Description

This project builds a long-horizon household item fetch-and-return system for a PAL Robotics TIAGo mobile manipulator in NVIDIA Isaac Sim. ROS 2 Nav2 handles room-scale navigation, while a LeRobot ACT visuomotor policy performs local approach and grasping.

Key Results

  • Trained on a combined 226-episode demonstration corpus.
  • Added residual PPO refinement over the frozen ACT policy; apartment-scene success improved from 83% to 91% under the matched evaluation protocol.

Links & References