Reinforcement Learning based Method for Autonomous Navigation of Mobile Robots in Unknown Environments

Abstract

The Reinforcement Learning is a subset of machine learning that deals with learning decisions from rewards given by the environment. The model classic reinforcement learning (RL) algorithms are usually applied to small sets of states and an action. However, in real applications, the state spaces are of a large scale and this will bring the problems in the generalization and the curse of dimensionality. In this research, authors integrate neural networks into reinforcement learning methods to generalize the value of all the states. The simulation results on the Gazebo software framework show the feasibility of the model proposed method algorithm. The robot can safely navigate an unprotected work environment and becomes a truly intelligent system with the ability to learn and adapt itself to the model.