Movement Generation And Task Recognition In Human Robot Collaboration

Close colloboration between robots and humans can give full play to their respective advantages and therefore significantly improve the working efficiency and job comfort.Movement generation and task recognition are the key to achieve natural human-robot colloboration,which howerver faces huge challenges.Due to the uncertainties in dynamic and unstructured working environment,traditional robot programming methods is still unable to adapt to such application.What's more,the traditional robot control algorithm is only suitable for simple and fixed task modeling,lacking of the ability of generating movements and recognition ability for complex tasks under dynamic environment.In order to improve the coordination and smoothness of human-robot collaboration,this paper mainly considers the movment generation and task recognition in human-robot collaboration.On the one hand,on the premise of human-centered,how can robots quickly and accurately identify human's intention.On the other hand,on the basis of human task recognition,how can robot accurately predict the interaction position that human may move to and generate the corresponding motion to complete the task cooperatively.Aiming to the problems above,this article adopts imitation learning to realize fast programming for robot.As we know that imitation is the reference that human quickly learn various skills in life.It is promising that the imitation learning mechanism apply to robotic system which can make it quickly learn movement skills.In addition,human handling of complex tasks tend to break it down into many steps and simple subtasks,as a guide,the robot's complex tasks can be described by the primitive serialization,through this way,it will effectively improve the diversity of tasks and generalization.In human-robot collaboration,the joint probability model of human-robot interaction are built,and then the probability distribution of robot and human is updated by observing the human motion state,so as to realize the recognition of human tasks and the generation of robot motion.In this paper,motion generation and task recognition in huamn-robot collaboration are systematically studied,The main research contents and achievements are as follows:(1)A human-robot collaboration system consisting of peripheral auxiliary equipment such as Baxter dual-arm robot,object recognition and pose estimation module and human motion capture module are built.The calibration of relative pose between devices is studied.Then,based on the system,the robot and human body movement information are collected.(2)A framework of FSM and DMPs is proposed to effectively describe the complex robotmanipulation tasks.DMPs is used to characterize and generalize motion trajectories in writing tasks to verify the motion coding ability and generalizability.Finally,the feasibility of the framework is verified through kitting experiment.(3)To overcome the limitation of the representation of motion,a Probabilistic Movement Primitives(ProMPs)is proposed to model the uncertainty of motion,which is the probability distribution of motion.After the motion model is established,the motion distribution is updated through the motion observation,and the corresponding motion can be inferred.Meanwhile,bayesian inference is used for task recognition.(4)Aiming at the human-robot physical interaction task,the joint probability modeling of human and robot motion is carried out by using ProMPs,so that the corresponding robot motion can be inferred by observing the human motion.This method is called the Interaction Probabilistic Movement Primitives(IProMPs).Aiming to increace the recognition rate,a dynamic update distribution method is proposed,error analysis is carried out,and corresponding human-computer interaction experiment is designed.