The Optimization Of The Mobile Robot-based Collaborative Order Picking Process Under The Constraints Of Human Factors

With the continuous improvement of warehouse automation,multi-robot systems are increasingly used in the warehouse environment due to their small space requirement,flexible demand response,and ability to work around the clock.The robotic mobile fulfillment system(RMFS),as an automated retrieval and storage system especially suitable for B2 C e-commerce order fulfillment,is widely used in the warehouse of Amazon,Walgreens,Zappos,Staples,JD,Tmall,Suning,and other famous e-commerce enterprises.The wide application of RMFS has dramatically increased the popularity of mobile robots in the warehousing environment and improved the efficiency of order picking in the warehousing system,but at the same time,it also brings extremely high physiological fatigue and stress to the human pickers who work with robots efficiently.At present,the relevant research of RMFS mainly focuses on the system structure design and operational strategy optimization,and few studies consider human factors.In logistics warehousing,where automation technology has largely replaced mechanical labor,higher requirements have been put forward for the order picking process that is hugely dependent on the pickers’ cognition,inference,decision-making,and operational capabilities.Human factors have become a key factor restricting the efficiency of the order picking system.Robot assignment decision-making and robot shelf transportation in the human-robots collaborative order picking,the two most essential components of the order picking process,are the prerequisite and guarantee for the smooth progress of the workstation pickers’ order picking operation,to determine the efficiency of RMFS.Therefore,this article focuses on the optimization of the mobile robot-based collaborative order picking process under the constraints of human factors,specifically as follows:Firstly,according to the subjective discomfort level of pickers in the process of RMFS order picking,human pickers’ quantitative model of the discomfort rating was established based on the Borg CR-10 evaluation scale,products location,products characteristics,and other factors.To solve the problem of robot assignment decision-making under different order picking tasks and different discomfort level distribution of pickers,a decentralized multi-agent reinforcement learning algorithm was proposed to train the autonomous assignment policy of mobile robots with the objective of balancing system efficiency and picker’s discomfort level.The simulations prove the effectiveness of the learned robot assignment policy in realizing the reasonable distribution of workload among the pickers.Secondly,the pupil diameter signal that objectively reflects the stress state of the picker is used,and the wearable physiological signal sensor is used to achieve real-time,non-sensing,and accurate detection of the physiological state of the picker.Considering the dynamic complexity of the system state and the uncertainty of human pickers’ stress level,based on the Value-Decomposition Network(VDN)multi-agent reinforcement learning algorithm,a reward function is constructed using the picker’s stress measurement and time cost,a decentralized robot autonomous assignment policy oriented to the picker’s stress level is obtained.The simulation proves that the proposed robot assignment policy can effectively reduce the stress duration of the pickers while ensuring the efficiency of order picking.Then,a robot assignment policy research framework oriented to picker fatigue-stress level management is proposed given the large order set and long-time products picking situation.Based on the use of pupil diameter signals to achieve stress detection,the picker’s heart rate signal detection is introduced to achieve the common detection of the picker’s fatigue-stress state.Based on the detected physiological state of the picker,the QMIX algorithm is used to obtain a real-time assignment policy based on the autonomous decision-making of the robot.A reasonable picking employee rest plan is obtained through the robot assignment and suspension of the assignment decision.The simulation test proves the learned robot assignment policy’s effectiveness in managing picker fatigue-stress levels.Finally,the tracking control process of robot shelf transportation trajectory in the human-robots cooperative order picking process is studied.Considering the model parameter perturbation,speed coupling,load variation,and other external disturbances encountered by mobile robots in shelf handling,the disturbance observer is introduced.Then,based on the proposed simplified generalized Kharitonov theorem,an image analysis method for determining the feasible region of the disturbance observer and speed controller parameters is presented.Simulations verify that the proposed method can ensure the effectiveness of trajectory tracking when the parameters of the robot model are subject to large perturbations and disturbances.In summary,this paper studies the optimization of the human-robots collaborative order picking process based on mobile robots under human constraints from the mobile robot assignment policy and the robot trajectory tracking control,so as to achieve the maximum adaptation to the pickers’ physiological state while ensuring the robotic mobile fulfillment system operation efficiency.It provides new research ideas and methods for the research on the human-robots collaborative order picking process under the e-commerce warehousing and provides an effective theoretical basis for improving the operating efficiency of the robotic mobile fulfillment system.