Research On Force Sensing And Compliance Control In Human Robot Collaboration

Collaborative robot(Cobot)has a wide range of application in military,spaceflight,medical treatment,nursing,industry and daily life after twenty years development.In some specific application tasks,such as human robot cooperation carry and assembly,limbs rehabilitation exercises or walking with robot assist device,the cobot should be requested to possess abilities of force sensing,intention recognition and making decisions in order to guarantee the safety,nature,smooth in collaboration with humans.However,because of force sensors like skin,variety and difficult estimation of cooperator model parameters and fixed roles limitation,estimation of contact information along whole manipulators,intention recognition and regulation of active control based on interaction force are difficult to perform,which bring about a series of problems,such as dangerous collaboration,inconvenient operation,cooperator's fatigue,vibration of robot and inefficiency.Therefore,a method of estimating whole body contact information based on proprioceptive sensors was presented in the pape,and compliance control techniques from intention recognition and roles assignment are researched and explored in order to realize robot manipulator tactile and compliance control.When a robot makes contact with external environments,the force data can provide important information for human robot security and control strategies,so the robot needs firstly to immediately sense the contact information.One dimension or multi-axes force sensor installed in robot's end is used to measure the interaction force in most human robot collaboration.However,lack of the manipulators tactile not only there are some potential safety hazards,but also the collaborative task can not be achieved through the operator touch the robot body,which limits the flexibility of the human robot collaboration system.Considering that the “skin-like” sensor is not mature and the structure of high integration joint is complex,we proposed a novel approach based on observation algorithm and optimization algorithm to solve the on-line estimation problem of the contact information.The proposed method transforms a complex detection problem of the any contact point on the robot body into a simple one-dimensional optimization solution with simple bound.The searching space determined by the momentum-based approach is limited within the length range of the contact arm.The combined algorithm is not only greatly increase the precision of estimation contact information,but also improve interactive real-time performance by the robot tactile sensation.The theoretical analysis and the experiments prove that the cobot can estimate the contact location and contact force from its any contact point on the robot body quickly,accurately and effectively without using external sensory systems.Endowing the robot with a certain amount of cooperation experience before tasks is essential to realize its active compliance control.A new intention estimation method for the collaborative robot was proposed through analysing the process of shaping the human experience and observing judgment strategies of an organism.After offline training based on radial basis function neural network,some cooperative skills of the robot can be obtained by the prediction model.During the online execution,the cobot can estimate the cooperator's intention according to the information of both the contact force from the human and the present state from itself.The proposed method can overcome the difficulty in establishing the model of the human robot cooperation based on traditional methods,especially for the complex and dynamic real-time movement model of human and the uncertainty impedance parameter values.Experiment results show that the cooperator' force is reduced while human robot synchronism motion is developed,so that the compliance of the cobot is improved greatly.For practical purposes,an adaptive roles allocation method based on reinforcement learning was proposed to solve dynamic roles adjustment.Firstly,the ideal performance of the common task was analyzed sufficiently during the different phases.A comprehensive cooperative performance model was created based on characteristics of the human and the robot,and was combied with the return function of the reinforcement learning.In the learning process,the roles of the human and the robot was adjusted based on the performance model,and the adjustment rules which can decide determine initiative for cooperative system were gradually formed,so that the system performance was continuously improved.The proposed method not only can help human robot cooperation system adapt to your new surroundings quickly,but also give full play to respective advantages of the cobot and the collaborator.It is meaningful to develop the human robot cooperation techniques no matter in theory or thereality.This thesis made an intensive study on estimation of robot manipulator contact information and compliance control strategies.This paper probes into these problems in terms of such robot contact information estimation,intention recognition,roles assignment and so on about human robot cooperation.The tactile sensation method was proposed and the old traditional strategies like modifying impedance parameters broken,which broaden the research on active compliance control methods.The research work in this thesis will be helpful for development of the robot that can think and make decisions,and will romote the cobot to intelligent way,which promotes physical human robot interaction process efficiency and has certain practical and theoretical values.