Further investigations on friction compensation using a neuro-genetic based hybrid framework

This paper presents further investigations into the use of a neuro-genetic based hybrid framework within a model-based friction compensation scheme in a closed-loop robotic system. The hybrid framework is composed of a number of neural network modules and a genetic algorithm module. The neural networks are used to perform a function approximation task while the role of the genetic algorithm is to search for an optimal combination between different neural structures during the generalisation process. This paper covers the modification on the genetic algorithm described in the previous work (Chaiyaratana et al. (2000)) to include two additional genetic operators: fitness scaling and diversity control operators. In addition, the search for an optimal combination between different neural structures is also extended to the case of the combination between radial-basis function networks, multilayer perceptrons and modular networks. The simulation results indicate that the friction compensation performance is further improved after the genetic algorithm and the search space has been modified. This helps to reveal the full potential of the hybrid framework in the friction compensation task