Research On Multi-AGVs Path Planning And Scheduling Technology Based On Reinforcement Learning

With the rapid development of smart manufacturing,the e-commerce industry and the intelligent warehousing industry,AGV has been more and more widely used as the most critical equipment in the the movement and transportation of materials and products.The application scale of AGV is increasing day by day,local congestion is easy to occur under road network without load balancing,which will lead to path conflict even deadlock.And the operating efficiency of the road network will be seriously affected.This paper mainly focuses on the load balancing of road network,conflict resolution and path planning technology for multi-AGVs applications.The main work includes:For the problem of AGV path planning under the rasterized road network scenario,using the AGV path planning algorithm based on Q-Learning,the path planning and obstacle avoidance of single AGV are realized.An improved reward mechanism is proposed to solve the dynamic path planning and conflict resolution problems for multiple AGVs in grid area.The effectiveness of the algorithm is verified through simulation.In order to fully ensure the operating efficiency of the road network and prevent congestion in large-scale AGV sorting system,the load balancing problem of the road network under largescale AGV application scenarios is studied,and a path planning algorithm combining load balancing and reinforcement learning is proposed.This method is based on the Q-Learning algorithm and takes the local congestion degree into consideration for the reward of the QLearning algorithm.Simulation are carried out using two-way random entrance and exit road network model.The results show that the proposed algorithm can effectively balance the load of road network.As the Q-learning based path planning algorithm for multi-AGVs converges slowly under more complex situations,a path planning algorithm based on DQN is proposed.The Q value of Q-Learning algorithm is fitted by neural network and the path planning efficiency for multiAGVs is improved.The software structure of the planning and scheduling system for multi-AGVs is discussed,and the key implementation technologies including database interaction and concurrent performance,and MQTT protocol communication capabilities are tested.