Online optimization of AGV transport systems using deep reinforcement learning

Kei Takahashi, Sogabe Tomah


In recent years, the distribution industry and the manufacturing industry have faced many challenges such as labor shortages and product diversification. For this reason, there is an attempt to automate the distribution and production process by using an automated guided vehicle (AGV) that can automatically carry a package to a predetermined place. However, in order to automate, there is a problem of how to optimize the movement path and transfer of AGV, and various studies have been conducted to solve it. In this paper, we propose a method to control multiple AGVs using deep reinforcement learning. To evaluate the deep reinforcement learning methodology, simulation experiments are performed to first train one model and then use the learned network to optimize another model. And simulation results show that the proposed method learns optimal or near-optimal solutions from past experience and provides superior performance in new environments.


Reinforcement Learning; Online Optimization; DQN; Transport systems; AGV

Full Text:



  • There are currently no refbacks.