Abstract: A new kind of multi-dimensional wrist force sensor applied to MotomamV3X robot is introduced. The characteristics of genetic algorithm (GA) and artificial neural networks (ANN) are compared. The operator of crossover and mutation for GA is improved. A kind of new dynamic modeling and compensation method is presented based on improved genetic algorithm for the proposed sensor. The dynamic modeling and compensation principle and the algorithms of improved genetic neural net-works (IGNN) are introduced and the dynamic model and compensation model are given for the proposed robot wrist force sensor. In this method, the dynamic model and compensa-tion model of wrist force sensor can be set up according to measurement data of the dynamic calibration, where the dy-namic model and compensation model parameters are trained by improved genetic neural network. So the method remains the global searching ability of GA and the simple structure and good robustness and self-learning ability of FLANN and can overcome FLANN’s shortcoming of easy convergence to the local minimum points and has fast network training speed and high modeling precision. Their effectiveness is verified by ex-periments and theoretical analysis.

Key words: Function link artificial neural networks, Dynamic compensation, Dynamic modeling, Genetic algorithm, Robot wrist force sensor