Calibration Of Wrist-mounted Six-axis Force/accelerometer Sensor And Compensation Of The Non-contact Force

Force control has been token into many applications in robotics,allowing robots to interact with the world and finish tasks compliantly.Especially in the industry,more and more sophisticated operations make it more valuable.The wrist-mounted force/torque(F/T)sensor is currently the most extensive method of F/T perception.Because the F/T sensor is attached between the robot and the load(or end-effector),its measurement always contains the external contact F/T and non-contact F/T,i.e.,gravity,inertial force,Coriolis force,centrifugal force and corresponding torque.The current methods for the compensation of the non-contact force require extremely expensive JR3 sensors(integrated 6 DOF F/T sensor and 6 DOF linear/angular accelerometer),or use high-precision Inertial Measurement Unit(IMU).The calibration accuracy of these sensors affect greatly the compensation error.As using the iterative method to calibrate,it is easy to fall into the local optimal solution.Aiming to avoid it and take a linear calibration,the research on the key calibration of sensors in the load dynamics compensation system has been invested.While the rotation between the F/T sensor and the robot can not be obtained through the dowel pins and there is a mechanical errors in some scenarios,a calibration method with an analytical solution is proposed.With a in-depth investigation of the load gravity compensation for the wrist-mounted F/T sensor,the rotation matrix between the F/T sensor and the robot is considered into the kinematics model.Based on the novel model,two calibration algorithms are proposed by the restricted and unrestricted orientations of the robot.Simulation and experiment verify the algorithm.The result show that it facilitates the use of force/torque sensors in more complex workshop,and obtains a more accurate gravity vector component of the non-contact force,about ±6%g.For the estimation of the acceleration and angular velocity of the load,the previous work use a high-precision Inertial Measurement Unit(IMU)and calculate the numerical rotation matrix between the IMU and the robot by using the iterative method.After investing deeply into the IMU calibration(especially,of the wrist-mounted IMU),a kinematics model of the wrist-mounted triaxial accelerometer is carried out,considering the correction and bias of the sensor,and the rotation matrix between the accelerometer and the robot.Based on this model,a linear calibration method is proposed to get an analytical solution.The numerical example and robot experiments verify the proposed calibration algorithm.Compared to the current iterative method,the linear calibration ensures the stability of the global optimal solution,and is easy to implement the online calibration with a much less calculation time.In the load dynamics compensation experiment,the two proposed calibration algorithms are successfully applied to calibrate the sensor.A method estimate the acceleration and angular velocity on-line by a based-quaternion Extended Kalman Filter(EKF),and identify the inertial parameters by Total Least Square(TLS).Based on these works,the non-contact force in the F/T sensor measurement is effectively compensated.In the experiment,the high-speed and violent motion is planned in the joint space for the robot,which causes a ±20N non-contact force.After the dynamics compensation,the max force error is 3 N,and the average error is 1.23 N.