A Master's Thesis Titled: Deep Q-learning based motion planning for mobile robot navigation

Abstract

This Master thesis implements two deep‐RL planners DDPG and TD3 for real‐time motion planning of a mobile robot in unknown environments. Agents generate continuous linear and angular velocity commands to reach randomized goals in ROS/Gazebo simulations on Pioneer 3‑DX and TurtleBot3 platforms. We train each agent across diverse obstacle layouts and analyze learning stability, sample efficiency, trajectory smoothness, and goal‐reaching success. The results offer actionable insights into integrating deep‐RL algorithms for robust, autonomous navigation in dynamic settings.