Distributed Computer Architecture and Fast Parallel Algorithms in Real-Time Robot Control

Control algorithms for accurate and fast moving robots require computations based on the complete dynamic model of the robot. These computations are time consuming due to the high degree of coupling between robot links and the nonlinear dependence of the robot's dynamic characteristics on orientation of the links. A method for processing the dynamic control equations in real-time by using parallel computation is described. The recursive Newton-Euler equations are distributed over multiple computing elements, one for each joint. Concurrency is achieved by substituting "predicted" values for the actual values of variables involved in the recursive equations. The method has been simulated and compared to other approaches to this problem such as simplification of the dynamic equations. The experimental results have shown that the errors introduced by prediction are relatively small and compare very favorably to other techniques. The computer architecture and algorithms presented offer a practical way to implement the control algorithms with relatively inexpensive computing devices.