PD-type control with neural-network-based gravity compensation for compliant joint robots

Abstract

Since the gravity terms depend only on the link positions in compliant joint robots, a neural-network-based gravity compensation scheme is conceived while the gravity model is unknown or is too complicated to be expressed explicitly. A PD-type control with this compensation is developed with the high-gain torque inner loop such that singular perturbation theory may be used to analyze the stability and passivity. Finally, three experiments are implemented to validate the effectiveness of the invented PD-type control with neural-network-based gravity compensation.