Advanced Robotics Navigation

SFU Robotics Course Project

Implemented an MCTS-DRL navigation framework in ROS2 (C++ and Python) for human-aware robot navigation. We incorporated an LSTM-based human action predictor to bias tree expansion toward likely future trajectories, and replaced random rollouts with a DDQN value function for more efficient planning, achieving 90.1% directional accuracy. Deployed end-to-end on a $10k QBot2e platform using VICON pose estimation and Cartographer SLAM, bridging simulation and real-world performance. Project website →