Force Estimation using a High-Fidelity Strain Wave Gear Model

In this paper, we show that the asymmetrical, and changing, efficiency of a robot joint has significant effect on the accuracy of sensorless force estimation methods. Our work is based on a high-fidelity strain wave gear model that includes gear meshing friction forces. The meshing friction cause the efficiency of the gear to depend on the load and on whether the joint is in forward drive or backward drive. The changing efficiency is not captured by standard robot dynamic models which assume rigid joints nor by most other high-fidelity models. We use the presented gear model for sensorless force estimation using a generalized momentum observer that allows the estimation of externally applied forces on the links of the robot. The external force estimation is pivotal for the control applied during kinesthetic teaching of collaborative robots.