Sensorless observation of the arm-side trajectory for iteratively increasing the path accuracy

Abstract:
High accuracy robot applications require knowledge of the end-effector or TCP position. Due to joint respective drivetrain elasticities, serial robots show deviations between the usually only measured motor side position and the resulting endeffector position. It would be possible to reconstruct the TCP position if the arm-side joint coordinates were measurable, which is usually not the case due to cost reasons. Therefore, one decentral Extended KALMAN-Filter for each joint is proposed in this paper. It estimates the unmeasurable arm-side joint space information. The state estimation is based on an identification with configuration-dependent consideration of the inertial and gravitational influences. A model of the complete coupled multi-body dynamics is utilized to compensate for couplings between different drive axes. Thus, despite simple decentralized observer models, a robust identification and state estimation is achieved. The proposed observer approach is evaluated by an experimental study for a 6-DOF industrial robot KUKAKR6-2. It exhibits a significant accuracy improvement compared to the typically used motor-side measurements.