Neural network-based variable impedance control of flexible joint robots

In this paper, a novel adaptive impedance control strategy for the flexible joint robot (FJR) is proposed. To simplify the controller design process, the singular perturbation technique is used to decompose the original high-order system into low-order subsystems. To reduce the mismatch of the system model, the neural network is used to estimate the friction and unknown system dynamic, where an improved optimal bounded ellipsoid (IOBE) algorithm is adopted to optimize the weight matrix of the neural network, which can fix the learning gain matrix vanishing or unbounded growth in traditional OBE algorithm. Different from traditional impedance controllers with fixed impedance parameters, in this paper, the variable stiffness and damping coefficients are used, which can maintain a fast response speed when the FJR is moving freely and can show more compliance characteristics when the FJR is interacting with the environment. The stability of the closed-loop system is proved via the Lyapunov approach and the effectiveness of the algorithm is verified by simulations.