Robust control of robots using mixed μ and neural network control

A method of designing a controller for manipulator arms which is robust to large payload variation is introduced. The method is to use neural networks to estimate some of the actual uncertain parameters only accurately enough to reduce their uncertain bounds to the point where a robust μ controller can be designed. The method is applied to design a controller for two-link robotic arms with unknown payload. The simulation results demonstrate that the μ controller and neural networks together can adapt to a wide range of variations of payloads. A μ controller without neural networks, developed here, provides robust performance over a limited range of changes in payload of about ±1 kg. By using a neural network estimator, this range is extended to ±5 kg