Robot Navigation with Reinforcement Learning

A TurtleBot3 Burger learns to drive to a goal through cluttered rooms, mazes and moving obstacles using only its 360-degree LiDAR - no map and no path planner. A PPO policy (Stable-Baselines3) is trained in a fast 2D simulator on a laptop CPU in 9 minutes and reaches the goal in 75.0 % of 400 unseen test episodes, against 55.5 % for a classic potential-field controller. The same observation, action and reward are implemented for ROS 2 Jazzy and Gazebo Harmonic.

Category: Robotics / Deep Reinforcement Learning / Autonomous Navigation / ROS 2 Simulation

CSE Project 29 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.