In situ calibration of joint torque sensors of the KUKA LightWeight robot using only internal controller data

Abstract

International audienceThe Kuka LWR is equipped with torque sensors mounted on the link side of the actuated joints. Each torque sensor is calibrated separately before it is mounted on the robot. This needs a second calibration at the last stage of the assembling of the robot in order to take into account the effect of the robot structure. This final in situ calibration is necessary to improve the accuracy of the estimation of the interaction wrench of the robot end-effector with its environment. However, the proposed calibration techniques are usually complicated, time-consuming, and must be carried out before assembling the sensors on the robot. In this paper, a simple and fast method for calibrating the sensors once they are assembled on the robot is presented. The method is based on the least squares solution of an over-determined linear system obtained with the robot inverse dynamic identification model in which are included the sensor gains. This model is calculated with sensor measurement and joint position sampled data available in the controller while the robot is tracking some reference trajectories without load on the robot and some trajectories with a known payload fixed on the robot. The method is experimentally validated on the Kuka LWR4+ but can be applied to any similar kind of robot equipped with joint torque sensors

Consistent kinematics and dynamics calibration of lightweight redundant industrial manipulators

“…For example, for the most complete Model 2 written originally with 136(= 8•17) IDM parameters (for 7 joints and a payload) the Symoro+ package [19] found eighteen non-identifiable parameters and suggested other sixteen to be regrouped resulting altogether in the set of base parameters with 102 elements. Forty four of them were later singled out by the manually supervised iterative procedure as essential [5]. Regrouping relations for base parameters were similar to ones reported in [17,Table III].…”

“…Works [5,16,17] are the closest to the dynamic calibration part of this study in terms of the model and signal processing algorithms used as well as experimental setup under consideration. However, it suggests to use non-optimized LSPB (trapezoidal velocity profile) calibration trajectories, which can significantly deteriorate estimation quality.…”