An Experiment in Neuro-Computed Torque Control of a Geared, DC Motor Driven Industrial Robot

Efficient control of robotic manipulators is an important industrial problem, and this paper proposes the use of artificial neural networks as a possible solution. The computed torque control scheme in particular is examined, and neural networks are used for the task of modelling the robotic manipulators. This scheme is applied to a two axis robot, whose feedback is corrupted by noise. The computation delay present in the system is subsequently accounted for by the use of neural network state predictors. The performance of the neuro-computed torque controllers is compared against a linear controller, and a comparison of other issues such as noise problems is included. Joint space and Cartesian space trajectory results demonstrate the superiority of the neural network-based nonlinear schemes over the linear controller. Key-Words: Robotics, Neural Networks, Computed Torque, Nonlinear Control, Noise Robustness