A comparison of techniques for modelling robot dynamics

An industrial robot can be programmed in two ways either, by writing the program online using teach programming or by generating the program off-line with the aid of a CAD based Off-Line Programming system(OLP). An OLP uses simulation to imitate the robot activities. This simulation is based on a model (for computing, a mathematical model) which sufficiently describes the robot kinematics, dynamics and the controller strategy. In general, analytical models are used. However, some investigators have proposed models based on neural networks. In this paper the advantages and disadvantages of neural network based models are explained and compared with general analytical modelling methods. Furthermore possibilities for the refinement and adaptation of the analytical model of the respective robot are discussed. This includes investigations regarding the applicability of automatic mechanisms for parameter identification etc. in order to automate model generation.