Research On Demonstration-based Task Learning And Autonomous Trajectory Generation Method Of Household Wheelchair Mounted Robotic Arm

Household wheelchair mounted robotic arm is a kind of service robot specially developed for the elderly and the disabled in today’s increasingly severe aging society.It can help users complete various daily household tasks and realize to live independently even without any care.Because this kind of robot considers both the fast mobility of the wheelchair and the flexible operation of the manipulator,it dramatically expands the activity space of the users and is particularly favored by the users.Due to the users’ physical function or cognitive deficiencies,it is difficult for them to flexibly control the robot to serve themselves.They hope the robot can autonomously provide “quick and easy to use” and “in line with the user’s operation plan” auxiliary assistance,and they still master the initiative of manipulation.Therefore,to meet the actual needs of the elderly and the disabled,this paper carries out the study of demonstration task learning and trajectory autonomous generation method to accomplish the daily household task autonomously.Based on this study,this paper develops the experimental robot platform and carries out the typical household tasks.Successfully obtaining the demonstration information of the daily household task is the prerequisite of carrying out the key-point-based demonstration information recording approach.Therefore,this paper first determines the typical tasks that need to be assisted by the robot and their task completion sequences based on daily household tasks.Simultaneously,to solve the problem that the demonstration information contains too many redundant operations or misoperations,this paper proposes a demonstration information recording approach based on key points to obtain the demonstration information and designs the dedicated interface to reduce the difficulty of operation.Additionally,this paper evaluates this approach from the quality of demonstration trajectories and demonstration process.Comparative experiment results of typical household tasks indicate that compared with the traditional direct demonstration information recording approach,the proposed approach can significantly reduce the users’ operation difficulty and mental burden,especially suitable for completing complex,tedious,multi-step daily household tasks.Trajectories of the daily household tasks are often complex and cumbersome and cannot be directly represented by a motion strategy.The above phenomenon requires the segmentation of the demonstration trajectory to extract its basic actions.In order to solve the problems of cumbersome operation,time-consuming,and inaccurate segmentation results caused by the manually segmentation method,this paper adopts the Beta Process Autoregressive Hidden Markov Model(BP-AR-HMM)to segment the demonstration trajectory of the daily household task.Especially,this paper firstly carried out a series of preprocessing on the demonstration trajectory such as Euler angler mutation process,improved recursive smoothing filter process,alignment process of multiple demonstration trajectories,and standardization processing;then,it adopts the BP-AR-HMM algorithm to segment the trajectory to extract the basic actions contained in the trajectory and lay the foundation for subsequent action learning and generalization.At the same time,it judges the properties of the obtained basic actions and establishes its corresponding demonstration task library.The segmentation results of the holding water glass task confirm the effectiveness and accuracy of the demonstration trajectory automatic segmentation approach proposed in this paper for daily household tasks.Reproducing the demonstration task in a new environment is the key of the robot to assist the elderly and the disabled to complete various daily household tasks successfully.In order to enable the robot to generate task completion trajectories that conform to the user’s operating habits in the new environment,this paper proposes a trajectory autonomous stitching generation approach based on the combination of an improved dynamic motion primitive and dynamic artificial potential field approach(Improved DMPs-DPF approach).Especially,the skillful actions are directly modified by the coordinate translation according to the target object’s position in the new environment;the transferable actions are modified by DMPs-DPF approach to carry out the appropriate learning expressions and generalize in the new environment.Additionally,this approach also considers the various shape obstacles that existed on the generalization path.Finally,this paper splices the generalized transfer actions and the translational skill actions following the order of the demonstration task to generate the complete trajectory in the new environment.Based on the above research,this paper builds the wheelchair-mounted robotic arm experimental platform.The platform is composed of JACO robotic arm,Xtion camera,Express electric wheelchair,and laptop.It can read the demonstration information of the specified task according to the user’s instruction,obtain the position information of the target object in the new environment with the help of the vision system,and automatically generate the complete reproduction motion trajectory.This paper carries out the eating task and holding water glass task to verify and evaluate the method proposed in this paper based on this experimental platform.Experimental results show that this method can control the robot to complete various daily household tasks and reduce the difficulty of task operation and save operation time.Compared with the manual operation mode,the autonomous operation can save about 45% of the time for complex and multi-step eating tasks,even for simple tasks of holding water glass task,this mode can save about 10% of the time.This phenomenon shows that the autonomous operation mode is suitable for application in the home environment for completing various complex and multi-step household tasks.