Research On Sensorless Human-Robot Interaction Control And Collision Detection Of Robotic Manipulator

As the requirements for human-robot collaboration in industrial settings are increasing,how to ensure that human operator and robot can safely share the working space and achieve smooth and convenient interaction has become a challenge for collaborative robot.In this dissertation,the small industrial manipulator is taken as the research object,and human-robot interaction goal of collision detection and lead-through programming without exteroceptive sensors is realized.Achieving human-robot coexistence and human-robot collaboration with a low-cost solution is not only economically valuable,but also can deepen theoretical research on human-robot interaction.To this end,this dissertation addresses the key issues in sensorless human-robot interaction control and collision detection including robotic manipulator dynamic parameter identification,interaction force observer design,collision detection and human-robot interaction control as follows.Accurate robotic manipulator dynamic model is the basis for interaction force estimation.For robotic manipulator dynamic parameter identification,a novel dynamic parameter estimation method combining robust linear regression and least squares variance component estimation is proposed.During the iterative estimation,outlier measurement data which do not fit the linear model are removed to improve the robustness of the estimation result to outlier data,and the best linear unbiased estimation of joint torque noise variance is obtained based on a reasonable assumption of joint torque noise variance structure.In order to ensure the physical meaning of the estimated parameters,physical feasibility constraint is also considered in parameter estimation.In addition,in order to reduce the influence of joint torque measurement noise on the parameter estimation,a more balanced excitation trajectory is obtained by optimizing joint sub-regression matrix based on the insight that robotic manipulator dynamic parameter identification result is the weighted sum of the independent identification result of each joint.The theoretical necessity of optimizing the sub-regression matrix is also given.Experimental results show that the proposed dynamic parameter identification and excitation trajectory design method can improve the accuracy of the identified dynamic parameter,effectively avoid the fitting accuracy degradation in the cross-validation.Compared with the traditional method,root-mean-square error of the fitted joint torque is reduced by up to 40.5%during cross-validation.In order to provide force information for human-robot interaction control and collision detection,disturbance observer needs to be designed to estimate the interaction force.Thus,a novel adaptive Kalman filtering algorithm is proposed to estimate the interaction force.Firstly,based on the perturbation analysis of the discrete-time algebraic Riccati equation,the criteria to determine whether adaptive Kalman filter can be applied to the system to be observed is given,and the interaction force observation system model which meets the requirement is given.Then,with theoretical analysis of the effect of noise covariance mismatch on the performance of Kalman filter,a novel adaptive Kalman filter algorithm is proposed combined with the Stein’s unbiased risk estimate.The effect of regularization term on the proposed adaptive Kalman filter is also discussed.Finally,the adaptive Kalman filter is employed to estimate the interaction force,and it is shown by simulation and experiment that the proposed algorithm can automatically balance the trade-off between noise amplification and tracking performance.Compared with several existing observers,the proposed method has the smallest estimation error overall.Load changes,friction coefficient variation and other factors make the robotic manipulator dynamics model inaccurate.The nonnegligible residual of the interaction torque observation adversely affects collision detection.Therefore,an adaptive filtering-based collision detection algorithm is proposed to cope with the model uncertainty.Firstly,a reasonably simplified linear model with time-varying parameters is given by analyzing the residual of the interaction torque estimation result.Then,in order to solve the time-varying parameter estimation problem under weak excitation condition,directionally forgetting recursive least squares method is used to avoid the estimation windup and parameter divergence problem of traditional adaptive filtering algorithms such as exponentially forgetting recursive least squares.Using the estimated timevarying parameter result,the filtering result of observed interaction torque residual and the adaptive dynamic detection threshold are obtained to realize the collision detection under model uncertainty.Human-robot interaction control and collision detection without external sensor are verified on the physical robotic manipulator.The interaction control of the manipulator without external sensor is achieved by combining the interaction force observation result and admittance control.Based on the robust performance optimization,optimal selection of admittance control parameters is given to realize the human-robot interaction control application like lead-through programming with continuous trajectory teaching in the operational space.Experimental result shows that the proposed method can achieve smooth and stable lead-through programming without force sensor.In collision detection verification,the estimated result of interaction force is used in the collision detection verification for both the cases where the dynamic model is relatively accurate and the model is inaccurate due to load.When the model is relatively accurate,the minimum collision force of 20 N is effectively detected with a simple threshold.When the model is inaccurate due to addition of a load to the manipulator,the proposed adaptive filtering collision detection method can overcome the disturbance of model uncertainty and achieve the detection of dynamic impact force.