Control of Nonlinear Elastic Joint Robots using Feed-forward Torque Decoupling

This paper proposes and evaluates the feed-forward decoupling of joint torque in elastic joint robots. The joint elasticities, captured by the third-order polynomial function, are considered together with the inverse manipulator dynamics so as to provide the required driving torque for the reference trajectories. Therewith decoupled joint actuators can be robustly controlled in a feedback manner for which a convenient proportional-derivative (PD) regulation is suficient. Using the inverse manipulator dynamics and joint model the reference value in the joint output (load) and not input (motor) space is provided to the feedback control of decoupled actuators. This allows compensating for the joint torsion and improve the load positioning accuracy without output sensors. The proposed control strategy is evaluated experimentally on a single elastic joint under gravity. We show an improved accuracy of the load positioning without adapting the underlying control loop and remarkable changes in the transient response.