Neural networks for modelling robot forward dynamics

Accurate identification of robot dynamics is essential for the implementation of advanced robot control algorithms. The nonlinearities present in a typical robot system make it difficult to use existing linear methods for this purpose. In this paper a new approach where feedforward neural networks are employed for robot system identification is presented. It is shown that the neural network model can accurately predict the behaviour of the robot system. Dynamic model of a two-link IBM robot is obtained using data generated by computer simulation, to illustrate the proposed method