Filtering Torque Ripples in Multi-Joint Systems: A Graph-Based Model Approach

Lightweight robots use harmonic drives and torque sensors to achieve high payload-to-weight ratios. Although torque controllers can mitigate the high reflected inertia inherent to these drives, their gains are limited by torque ripples. These periodic vibrations, occurring at twice the motor frequency, are amplified by the controller’s feedback loop, leading to instability with improperly selected gains. Previous research focused on filtering these ripples in a joint-wise fashion. However, we show that this joint-wise approach is insufficient during multi-joint motion, as ripples propagate through the robot’s kinematic chain. To address this, we propose a time-varying band-pass filter that accounts for ripple propagation using a graph-based representation of the robot’s instantaneous inertial properties. We validate our approach on a KUKA LBR Med 7 under co-manipulation, where, by utilizing the filtered signal, the apparent rotor inertia of the system is either halved or increased by five times. In the latter case, this approach halves the effects of external forces without materially changing the stability properties of the impedance controllers.