Q学習による多数AGVの自律搬送に関する研究

抄録

This paper reports on simulation results of multiple AGVs transportation in a virtual factory. AGVs drive on their ways autonomously. To do so, AGV learns paths between machining cells, and automated warehouse and machining cells by use of Q-learning before simulation. The fast learning methods, which take a consideration in failure learning, are proposed and the efficient learning method is found out. After the paths are learnt, transportation simulation for workpieces by autonomous AGV driving is performed. Results show that AGVs can transport the workpieces from the warehouse to machining cells and transport them back to the warehouse without any pre-programming and instructions, if enough number of AGVs are set in the factory. However, it is observed that deadlock phenomenon is observed very few times during transportation.