A hybrid intelligent active force controller for articulated robot arms using dynamic structure network

Active Force Control (AFC) is known to offer a feasible solution to control efficiently a robot operating at highspeed operation. This method is proven to be stable and robust against internal and external disturbance and has been successfully in various experimental works. In this regards, the main issue of AFC method is the estimation of the inertia matrix of the robot arm. In previous work neural network is proposed to estimate the inertia matrix. In this approach, the off-line training of neural network requires. systematic and effective data preparation scheme. In addition, how to design the neural network is another important issue. This paper proposes yet another methodology to address the two issues stated above. In this approach, the All estimation of inertia matrix is made based on priori knowledge of robot dynamic model. On the other hand, the optimal design of neural network is autonomously determined through the off-line learning phase using Growing Multi-Experts Network (GAN).