Dispatching and Path Planning of Automated Guided Vehicles based on Petri Nets and Deep Reinforcement Learning

A formal approach is proposed for scheduling a team of automated guided vehicles (AGVs). First, a team of AGVs and their environment are modeled as a place timed Petri net (P-timed PN). Second, a method is presented to design controlling structures to avoid collisions and reduce deadlocks of P-timed PNs. Third, a multi-AGV path planning problem is formulated as a Markov decision process, and a neural network is trained based on a corresponding reinforcement learning with the PN model to estimate the action reward function for a given Multi-AGV system. Finally, an approach is obtained to dispatch transportation tasks among and to plan routes for AGVs according to rewards calculated by the neural network. An example is utilized to illustrate the proposed methods.