Abstract
A routing method is suggested for automated guided vehicles using a reinforcement learning technique. This paper focuses on an algorithm called Qlearning that can acquire optimal routing strategies from delayed rewards, even when the agent has no prior knowledge of the effects of its actions on the environment. In manufacturing shops, there is a high possibility that vehicles on the way to the destination will experience unexpected delays due to interference from other vehicles. Thus, routes of the shortest travel distance are not necessarily the shortest in travel time. This paper discusses how the Q-learning technique can be applied to the routing problem. A numerical experiment was performed to evaluate the performance of the rules obtained from the learning process and the speed of the convergence of an objective value. The performance of the learning-based rules was compared with that of the shortest distance rule.
