The unmanned warehouse dispatching system of the ‘goods to people’ model uses a structure mainly based on a handling robot, which saves considerable manpower and improves the efficiency of the warehouse picking operation. However, the optimal performance of the scheduling system algorithm has high requirements. This study uses a deep Q‐network (DQN) algorithm in a deep reinforcement learning algorithm, which combines the Q‐learning algorithm, an empirical playback mechanism, and the volume‐based technology of productive neural networks to generate target Q ‐values to solve the problem of multi‐robot path planning. The aim of the Q‐learning algorithm in deep reinforcement learning is to address two shortcomings of the robot path‐planning problem: slow convergence and excessive randomness. Preceding the start of the algorithmic process, prior knowledge and prior rules are used to improve the DQN algorithm. Simulation results show that the improved DQN algorithm converges faster than the classic deep reinforcement learning algorithm and can more quickly learn the solutions to path‐planning problems. This improves the efficiency of multi‐robot path planning.

Public health emergencies have brought great challenges to the stability of the e-commerce supply chain. Demand forecasting is a key driver for the sound development of e-commerce enterprises. To prevent the potential privacy leakage of e-commerce enterprises in the process of demand forecasting using multi-party data, and to improve the accuracy of demand forecasting models, we propose an e-commerce enterprise demand forecasting method based on Horizontal Federated Learning and ConvLSTM, from the perspective of sustainable development. First, in view of the shortcomings of traditional RNN and LSTM demand forecasting models, which cannot handle multi-dimensional time-series problems, we propose a demand forecasting model based on ConvLSTM. Secondly, to address the problem that data cannot be directly shared and exchanged between e-commerce enterprises of the same type, the goal of demand information sharing modeling is realized indirectly through Horizontal Federated Learning. Experimental results on a large number of real data sets show that, compared with benchmark experiments, our proposed method can improve the accuracy of e-commerce enterprise demand forecasting models while avoiding privacy data leakage, and the bullwhip effect value is closer to 1. Therefore, we effectively alleviate the bullwhip effect of the entire supply chain system in demand forecasting, and promote the sustainable development of e-commerce companies.

Robotic mobile fulfillment system (RMFS) is a new type of parts‐to‐picker order picking system, where robots carry inventory pods to stationary pickers. Because of the difference in working mode, traditional storage assignment methods are not suitable for this new kind of picking system. This paper studies the storage assignment optimization of RMFS, which is divided into products assignment stage and pods assignment stage. In the products assignment stage, a mathematical model maximizing the total correlation of products in the same pods is established to reduce the times of pod visits, and a scattered storage policy is adopted to reduce system congestion. A heuristic algorithm is designed to solve the model. In the pods assignment stage, a model is established minimizing the total picking distance of the mobile robots considering the turnover rate and the correlation of pods as well as the workload balance among picking corridors. A two‐stage hybrid algorithm combining greedy algorithm and improved simulated annealing is designed to solve the model. Finally, a simulation experiment is carried out based on the historical order data of an e‐commerce company. Results show that the storage assignment method proposed in the paper significantly improves the efficiency of order picking.

A Robotic Mobile Fulfillment System (RMFS) is a new type of parts-to-picker order fulfillment system where multiple robots coordinate to complete a large number of order picking tasks. The multi-robot task allocation (MRTA) problem in RMFS is complex and dynamic, and it cannot be well solved by traditional MRTA methods. This paper proposes a task allocation method for multiple mobile robots based on multi-agent deep reinforcement learning, which not only has the advantage of reinforcement learning in dealing with dynamic environment but also can solve the task allocation problem of large state space and high complexity utilizing deep learning. First, a multi-agent framework based on cooperative structure is proposed according to the characteristics of RMFS. Then, a multi agent task allocation model is constructed based on Markov Decision Process. In order to avoid inconsistent information among agents and improve the convergence speed of traditional Deep Q Network (DQN), an improved DQN algorithm based on a shared utilitarian selection mechanism and priority empirical sample sampling is proposed to solve the task allocation model. Simulation results show that the task allocation algorithm based on deep reinforcement learning is more efficient than that based on a market mechanism, and the convergence speed of the improved DQN algorithm is much faster than that of the original DQN algorithm.

Robotic Mobile Fulfillment System (RMFS) is a new type of parts-to-picker order picking system and has become the development trend of e-commerce logistics distribution centers. There are usually a large number of tasks need to be allocated to many robots and the picking time for e-commerce orders is usually very tight, which puts forward higher requirements for the efficiency of multirobot task allocation (MRTA) in e-commerce RMFS. Current researches on MRTA in RMFS seldom consider task correlation and the balance among picking stations. In this paper, a task time cost model considering task correlation is built according to the characteristics of the picking process. Then, a multirobot task allocation model minimizing the overall picking time is established considering both the picking time balance of picking stations and the load balance of robots. Finally, a four-stage balanced heuristic auction algorithm is designed to solve the task allocation model and the tasks with execution sequence for each robot are obtained. By comparing with the traditional task time cost model and the algorithm without considering the balance among picking stations, it is found that the proposed model and algorithm can significantly shorten the overall picking time.