RBF neural network with genetic algorithm optimization based sensitivity amplification control for exoskeleton

To improve performance of sensitivity amplification control (SAC) for exoskeleton robot, genetic algorithm (GA) and RBF neural network was combined to obtain accurate dynamic model of exoskeleton robot online. Parameters of center vector and base width of RBF neural network were obtained by GA optimization, and online RBF weights learning process was constructed to obtain estimation matrix parameters of dynamics system, finally, SAC control law was deduced. Simulation results showed that the RBF network optimized by GA could learn exoskeleton dynamics model parameters online. Based on the learned model, the SAC could achieve more precise human trajectory tracking where tracking error and human-robot interaction force converged to the small neighborhood of zero simultaneously compared with those without optimization. The proposed RBF neural network with GA optimization which learned dynamics model parameters online for exoskeleton robot dynamics model could achieve highly accurate trajectory following for SAC, ultimately realize cooperative movement between human and exoskeleton.