Neural Learning Enhanced Variable Admittance Control for Human-Robot Collaboration

In this paper, we propose a novel strategy for human-robot impedance mapping to realize an effective execution of human-robot collaboration. The endpoint stiffness of the human arm impedance is estimated according to the configurations of the human arm and the muscle activation levels of the upper arm. Inspired by the human adaptability in collaboration, a smooth stiffness mapping between the human arm endpoint and the robot arm joint is developed to inherit the human arm characteristics. The estimation of stiffness term is generalized to full impedance by additionally considering the damping and mass terms. Once the human arm impedance estimation is completed, a Linear Quadratic Regulator is employed for the calculation of the corresponding robot arm admittance model to match the estimated impedance parameters of the human arm. Under the variable admittance control, robot arm is governed to be complaint to the human arm impedance and the interaction force exerted by the human arm endpoint, thus the relatively optimal collaboration can be achieved. The radial basis function neural network is employed to compensate for the unknown dynamics to guarantee the performance of the controller. Comparative experiments have been conducted to verify the validity of the proposed technique.