Robustness of Interaction Parameters Identification Technique for Collaborative Robots

Abstract

The work focuses on accuracy of the identification procedure allowing to estimate the force and its application point in human-robot physical interaction. It is assumed that the desired parameters are estimated using data obtained from internal torque sensors embedded in the robot joints as well as 3D geometric models describing the robot link surfaces. In practice, the measurement data are corrupted by the noise, which causes identification errors. To evaluate these errors, the relevant covariance matrix is obtained assuming that the measurement noise is presented as unbiased and independent but not identically distributed random values. Based on the relevant analysis, enhancement of the existing method was proposed, which improves the identification accuracy and its robustness with respect to the measurement noise. Particular attention is paid to singular cases arising when the estimated force action line intersects one or several joint sensor axes or does not intersect the robot body at all. The efficiency of the developed techniques was confirmed via simulation and experimental studies with the KUKA iiwa robot.