Control of Industrial AGV Based on Reinforcement Learning

Abstract

Automatic Guided Vehicles (AGV) suffer degradation in their electro-mechanical components which affect the navigation performance over time. The use of intelligent control techniques can help to alleviate this issue. In this work a new approach to control an AGV based on reinforcement learning (RL) is proposed. The space of states is defined using the guiding error, and the set of control actions provides the reference for the velocities of each wheel. Two different reward strategies are implemented, and different updating policies are tested. Simulation results show how the RL controller is able to successfully track a complex trajectory. The controller has been compared with a PID obtaining better results.