A Fast Computational Scheme for Dynamic Control of Manipulators

Abstract

The complex dynamics of manipulators make design and implementation of dynamic control laws for them very difficult. Today's industrial robots employ dynamic control laws which neglect the complex dynamics, thus leading to less than optimal response. The computational bottleneck of many advanced control schemes is an algorithm for computation of the actuator torques (forces) required to produce desired joint accelerations, for a given set of joint velocities and angles (displacements). The scheme described here, when implemented on a single processor, performs this computation about five times faster than the recursive Newton-Euler algorithm. When implemented on two processors, it is about ten times faster. The model used in the computation partitions the manipulator into two parts, is general enough to fit most practical manipulators, and uses no approximations.