Control of flexible joint robot manipulators using a combined controller with neural network and linear regulator

Abstract

This paper addresses the issue of dynamic trajectory tracking control of flexible joint robot manipulators. A new control strategy based on the technologies of linear feedback and neural network is proposed. It is well known that linear feedback controllers are simpler and easier to design, and neural networks have learning ability. These two advantages are utilized in order to deal with the control problems arising from non-linearity and dynamic instability of the flexible joint robot. The linear feedback control scheme is designed using optimal control approach, and the neural network is then designed based on the BP method and added to the control system. There are two different ways to add the neural network into the control system: being an additional regulator in parallel with the linear feedback controller, and being a dynamic compensator with some function of regulator. This leads to two different control systems. Performance of these two control systems are compared based on simulation results. Experiments are carried out using a two-link flexible joint robot as a test bed. The results demonstrate the effectiveness and usefulness of the proposed control method.