Neural network based robust adaptive control for a variable stiffness actuator

In this paper we present a robust adaptive controller based on a neural network (NN) for a variable stiffness actuator (VSA). The controller is able to independently set the mechanical stiffness and position at the joint shaft to guarantee robustness with respect to slowly time-varying and unmodeled friction coefficients affecting the dynamics of the actuator. The lumped uncertainties of the VSA including unmodeled dynamics are considered and approximated by a simple NN so that the controlled system is asymptotically stable, and remains effective while process conditions vary. To cope with the reconstruction error of the NN, a sliding mode like additional robust control term is introduced. The proofs for the uniformly ultimately bounded (UUB) and uniform asymptotic (UAS) stabilities for the closed-loop system are provided in detail via Lyapunov theory. Simulation and experimental results are reported in support of both validity and performance of the proposed approach.