State Recognition and Reinforcement Learning for Two-Wheel Mobile Robot

The autonomous robots recognize surrounding environments using sensors, make judgments based on algorithms, and control actuators such as motors. This paper presents a strategy to deal with unpredictable environmental changes of line following two-wheel robots. We define the running conditions of the robot from the light intensity value obtained from the sensor. Then, appropriate change amounts of the running parameters are derived by means of policy gradient reinforcement learning. With this function, the robot was able to properly provide the ability to dynamically respond to unexpected environmental changes.