Jincao Zhang MSc Dissertation 2015/16

Reinforcement Learning withRobotics:Analysis andImplemetation

Supervised by E.Vasilaki

Abstract

This thesis focuses on the demonstration of Reinforcement Learning in Robotics by analyzing and implementing two mobile robot agent navigation task, with the consideration of simplicity, applicability, and flexibility. The mobile agent is designed as an environmental free model (i.e., do not require any model of the physical environment)for accomplishing some simple obstacle avoidance tasks and target finding.Q-learning is chosen as the main learning algorithm since it is a simple, effective and most-known RL algorithm. The subsumption architecture is applied for multi-layer behavior coordination.