Autonomous Navigation for Exploration of Unknown Environments and Collision Avoidance in Mobile Robots Using Reinforcement Learning

This paper proposes the use of reinforcement learning to provide navigation and adaptation skills to a mobile robot in unknown environments. The use of sensor information is developed instead of the spatial location of the robot with the aim to perform the navigation and exploration process. The use of the Q-learning algorithm is proposed, however, the states are established based on the information coming from the sensors. A reward policy is applied to focus on guide the robot away from obstacles and allowing the exploration of unknown environments. Additionally, an exploration policy is generated which collaborates in the adaptation process of the robot. The complete system is evaluated using a robot simulation environment known as V-Rep. The results show that the robot learns to navigate avoiding to collide with obstacles, besides presenting skills to explore unknown environments.