| The rise of the digital economy and platform economy has placed higher demands on the service levels of modern logistics systems,particularly warehouse systems.As a core component of the warehouse system,the picking system is critical in determining the operational efficiency of the warehouse system,improving the service capabilities of the logistics system,enhancing customer satisfaction,and increasing the competitiveness of the platform.In recent years,in order to meet the requirements of large-scale,high-level,and flexible picking,mobile robots represented by KIVA have been widely applied in intelligent picking systems.The complexity of intelligent picking systems with mobile robots as task executors has significantly increased compared to traditional automated picking systems.Improving picking efficiency has become the focus and difficulty of system optimization.Order scheduling and the coordinated optimization of multiple robot paths are the key factors in improving picking efficiency.However,related research in this area is not yet sufficiently in-depth.In terms of picking modes,most research focuses on optimizing the KIVA system,with insufficient theoretical research on optimizing multi-load mobile robot systems.In terms of order scheduling,most research is limited to task assignment problems,with a lack of research on order sorting and assignment problems.In terms of multi-robot path planning,there is a lack of efficient algorithms applicable to large-scale,multi-round,low-cost,and scalable robot picking systems.Furthermore,there is no research on the coordinated optimization of order scheduling and multi-robot path planning.This article focuses on the intelligent warehouse system and conducts research on the coordinated optimization of order strategy and path planning in a multi-load mobile robot picking system.By optimizing the sorting and distribution of orders,the feasibility and optimality of the order scheduling plan can be improved.Through multi-robot path planning,the robot can complete picking tasks quickly without any conflicts in the path.Additionally,the coupling between order scheduling and path planning is considered,and the comprehensive optimization of the picking process is achieved through order-path coordinated optimization.The main contents and innovative points are as follows:(1)Regarding the problem of order scheduling in intelligent warehouse outbound tasks with unknown completion time,a mathematical programming model is established with the objective of minimizing the maximum completion time.A heuristic order scheduling algorithm based on estimated completion time is proposed.In the algorithm,a shortest path estimation strategy based on Manhattan distance is first designed.Then,to optimize order sorting,three heuristic strategies are designed based on the longest completion time rule for the first round task heuristic sorting,the shortest completion time rule for the same area task heuristic sorting,and the target picking station determination.In addition,a first round random sorting strategy and a subsequent same area priority assignment strategy are proposed to optimize task assignment,and a cross-area assistance strategy under non-uniform distribution of orders is designed.Three sets of simulation experiments are designed to test the performance of the order scheduling algorithm.The algorithm example verifies the feasibility of the algorithm,and the comparison experiments with multiple heuristic strategies verify the effectiveness of the algorithm.The algorithm comparison experiment verifies the superiority of the algorithm.(2)For the centralized and collision-free path planning problem of a multi-AGV(Automated Guided Vehicle)system in the picking process of an intelligent warehouse,a mathematical programming model is established to minimize both the maximum completion time and the ratio of invalid paths,based on the "box-to-person" mode of multi-AGV systems instead of the "shelf-to-person" mode of KIVA systems.A path planning algorithm called PCBS(Priority-based Conflict-based Search)is proposed.In this algorithm,a high-level conflict resolution acceleration strategy based on priority rules is proposed.Then,a dynamic programming algorithm is designed to optimize the execution sequence of multiple tasks in a single round of a single multi-AGV.Finally,considering the robot turning penalty in practice,an improved A*low-level path search algorithm is designed.To verify the effectiveness of the "box-to-person" picking mode,a comparison experiment between the MLAMR mode and the KIVA mode is designed.To test the performance of the multi-AGV path planning algorithm,a comparison experiment between PCBS and CBS algorithms and their various improved algorithms is designed.The experimental results show that the "box-to-person"picking mode of the multi-AGV system is effective,and the proposed PCBS algorithm has superiority over the compared algorithms on large-scale problems.(3)Aiming at the problem of distributed collision-free path planning for multi-AGV clusters in the outbound process of intelligent warehousing systems,a mathematical model based on Markov decision process is established.A CJ-MAPPER algorithm based on improved evolutionary reinforcement learning multi-agent path planning is proposed.The algorithm is improved in four aspects.In order to improve the observation efficiency of the robots,an observation space splitting method is designed.To reduce the state space while considering the actual situation of the warehousing map,the robot action space is simplified to four.In terms of reward function,a novel reward function consisting of six reward items is designed.In order to improve training efficiency,an evolutionary training strategy based on crossover and mutation is designed.Simulation experiments show that the proposed algorithm can achieve low-cost and rapid optimization effect.(4)Regarding the problem of multi-robot distributed path planning in the outbound process of intelligent warehousing systems,a mathematical model based on Markov decision processes is established,and an improved distributed path planning algorithm CJ-MAPPER based on evolutionary reinforcement learning for multi-robot clusters is proposed.The algorithm is improved in four aspects.In order to improve the observation efficiency of robots,an observation space splitting method is designed;in order to reduce the state space and consider the actual situation of the warehouse map,the robot action space is simplified to 4 actions;a novel reward function consisting of 6 reward items is designed;and a training strategy based on crossover and mutation is designed to improve training efficiency.Simulation experiments show that the proposed algorithm can achieve low-cost and rapid optimization results. |
