A Collaborative Control Method For Robot Wheelchairs Based On Head Gesture Interaction Skill Learning

Human-chair collaboration,as one of the key technologies in the field of human-robot interaction of robot wheelchairs,has attracted widespread attention from many research institutions at home and abroad in recent years.In the current research on human-chair collaboration,most of the assisted driving of robot wheelchairs treats the wheelchair as a mobile robot and directly adopts the SLAM(Simultaneous Localization and Mapping)technology and navigation technology of mobile robots,transporting the user from one place to another like a loaded object,without considering the user’s comfort during the whole moving process,nor absorbing human driving experience,nor generating human-like driving behavior.The driving of robot wheelchairs,in addition to considering safety and efficiency,also needs to pay attention to the user’s comfort.Only with a good comfort experience,can the user regard it as part of his body and give full trust during the moving process.This requires establishing a balance between safety,efficiency and comfort,which is often difficult to achieve without human experience data as a reference.On the other hand,patients will inevitably experience some muscle fatigue when operating robot wheelchairs for a long time,which may affect the control of robot wheelchairs and pose certain safety risks;and when robot wheelchairs are in fully autonomous driving mode,they will provide excessive assistance to patients,which will also cause patients to feel depressed due to lack of control experience.In view of these problems,based on the robot wheelchair interaction control based on head pose estimation,this paper carries out a series of research on the collaboration mode between human and robot wheelchair by learning the driving operation skills of patients.The specific research contents are as follows:(1)A method for learning robot wheelchair operation skills based on head pose interactionFor the collected driving demonstration data,beta process autoregressive hidden Markov model and dynamic time warping are used for skill segmentation and alignment;then Gaussian mixture model,Gaussian mixture regression and dynamic motion primitive are jointly used for processing and learning to extract skill regularity and variability,and further perform skill representation and generalization to obtain driving sub-skills and store them in DMP skill library;finally,the sub-skills in the skill library are generalized and invoked according to the task requirements to obtain the overall driving skill of the task,and Gaussian process regression is used to regress it through the driving trajectory to realize the reproduction of the operation skill from the position level.(2)A method for human-machine collaboration of wheelchair based on head pose interaction skill learning and fatigue perceptionA shared control system based on muscle fatigue factor is designed.The surface electromyography signals of the corresponding muscles are collected from the user.A muscle fatigue quantification formula is proposed and used to calculate the user’s fatigue level.Combined with the supportive shared control system,when the fatigue quantification value exceeds the set fatigue switching point,the user’s control authority is switched.The robot wheelchair uses the learned driving skills for autonomous control.When the user’s fatigue recovers,the wheelchair control authority is regained.(3)A method for human-machine collaboration of robot wheelchair based on decision-type shared learning controlA decision-type shared control system is designed for robot wheelchairs.By collecting real-time radar sensor and odometer data,and combining with user muscle fatigue,the user’s control ability is comprehensively evaluated,and the dynamic adjustment of human-machine control command weight coefficient is realized,so that the user and the learned operation skill can jointly control the robot wheelchair in real time.