Navigation of Ground Robots with Reinforcement Learning

Reinforcement learning (RL) is a machine learning technique that enables an agent to learn optimal behaviors within an environment through a process of trial and error. This is achieved by the agent receiving rewards or punishments based on its actions. In this study, we compare the performance of two RL algorithms, SARSA and Q-Learning, in a simulated environment using the Gazebo Simulator. The goal of the simulation is to navigate a ground robot towards pre-defined goals. By manipulating various training parameters, we investigate the impact on learning speed and robot behavior. To ensure meaningful comparisons, we vary the navigation goal and the complexity of the simulation environment. Through extensive simulations, our results highlight the effectiveness of RL-based navigation for ground robots and offer insights into the influential parameters that significantly affect optimal navigation performance. It underscores the importance of algorithm selection and parameter optimization in achieving optimal navigation performance.