Parallel processing in optimal control of robot drives

There is a class of drives which need to operate over a very wide range of disturbances and with some unknown variation in system parameters. A typical joint drive for an industrial robot is an example. The established design methodology of linear optimal controllers derived off-line, is not very well suited due to parameter variations and optimization of performance over a fairly wide and variable work volume of the robot arms. The real-time algorithms derived in the paper are particularly suitable for such drives because of their relative simplicity in terms of computing requirement and their applicability even when the system parameters are time-varying or not known precisely. The analytical derivations in the paper cover four algorithms: state estimation using Kalman theory, optimal feedback using estimated state variables and dynamic programming, adaptive algorithms and optimal inferential controller. Implementation of these algorithms on transputers with the specific aim of studying their load torque disturbance abilities forms the core of the paper. These algorithms were applied for real-time control of a small permanent magnet DC motor to validate both the mathematical concepts and their numerical accuracy