Adaptive Neural Trajectory Tracking Control for Flexible-Joint Robots with Online Learning

Collaborative robots and space manipulators con-tain significant joint flexibility. It complicates the controldesign, compromises the control bandwidth, and limits thetracking accuracy. The imprecise knowledge of the flexiblejoint dynamics compounds the challenge. In this paper, wepresent a new control architecture for controlling flexible-jointrobots. Our approach uses a multi-layer neural network toapproximate unknown dynamics needed for the feedforwardcontrol. The network may be viewed as a linear-in-parameterrepresentation of the robot dynamics, with the nonlinear basisof the robot dynamics connected to the linear output layer. Theoutput layer weights are updated based on the tracking errorand the nonlinear basis. The internal weights of the nonlinearbasis are updated by online backpropagation to further reducethe tracking error. To use time scale separation to reduce thecoupling of the two steps – the update of the internal weightsis at a lower rate compared to the update of the output layerweights. With the update of the output layer weights, ourcontroller adapts quickly to the unknown dynamics changeand disturbances (such as attaching a load). The update of theinternal weights would continue to improve the converge of thenonlinear basis functions. We show the stability of the proposedscheme under the “outer loop” control, where the commandedjoint position is considered as the control input. Simulationand physical experiments are conducted to demonstrate theperformance of the proposed controller on a Baxter robot,which exhibits significant joint flexibility due to the series-elastic joint actuators.