Jérôme Kirchhoff1, Oskar Von Stryk2
10:30 - 10:45 | Mon 25 Sep | Room 207 | MoAT11.1
Accurate velocity estimation is an important basis for robot control, but especially challenging for highly elastically driven robots. These robots show large swing or oscillation effects if they are not damped appropriately during the performed motion. In this paper, we consider an ultra lightweight tendon driven series elastic robot arm equipped with low-resolution joint position encoders. We propose an adaptive Kalman filter for velocity estimation that is suitable for these kinds of robots with a large range of possible velocities and oscillation frequencies. Based on an analysis of the parameter characteristics of the measurement noise variance, an update rule based on the filter position error is developed that is easy to adjust for use with different sensors. Evaluation of the filter both in simulation and in robot experiments shows a smooth and accurate performance, well suited for control purposes.