A Reinforcement Learning Approach to Autonomous Speed Control in Robotic Systems
Model-free reinforcement learning techniques have been successfully used in diverse robotic applications. In this thesis, we Implement the Q-learning algorithm as one of the most used model-free algorithms to find an optimal control signal for driving fast running trains on fixed tracks without flipping over or derailing. We examine the performance of the human driver and compare the results of reinforcement learning based controller to human driver performance. To test the proposed algorithm, a complete hardware and software testbed has been designed. We conclude that in simple tasks, human drivers perform identical to reinforcement learning algorithm, but in more complicated tasks, our reinforcement learning implementation performs better.