Q-Learning based Path Planning and Predictive Control for the Navigation of a Mobile Robot

Abstract

Our work aims to find the optimal path to enable a mobile robot to navigate from a starting point to a destination in a known environment, while avoiding obstacles. To achieve this goal, we started by studying and implementing the Model Predictive Control (MPC) framework in the first phase. Then, in a second phase, we explored various state-of-the-art planning algorithms, including Reinforcement Learning approaches. Among the latter, we studied and implemented the Q-Learning algorithm to perform the path planning according to the simulated scenarios. Ours simulations were conducted both using the Matlab environment and the MATLAB-ROS interface along with the Gazebo simulator. The results we obtained were highly reliable.