The dynamic equations of a manipulator are modeled by a system of second-order nonlinear differential equations with its kinematic and dynamic parameters. On the basis of this model, the manipulator can be controlled in a strict way by compensating all dynamic torques/forces consist of inertial, centrifugal, Coriolis, and frictional effects. However, since these dynamic torques/forces are highly complex functions of joint positions and velocities, the computational burden for evaluating these torques is significant. Further, the Cartesian-space control, in which positions and velocities are controlled directly in Cartesian work space, requires considerable amount of additional computations for inverse kinematics. Hence it has been difficult to implement the realtime model-based control with conventional commercially available microprocessors.
This paper presents a real-time implemetation scheme of Cartesian-space nonlinear feedback control based on a new parallel computation scheme called Resolved Newton-Euler algorithm. Some experiments on the basic three joints of the PUMA 560 manipulator are also reported by using a parallel processing system with multiple microprocessors. A sampling period of 0.79 msec is achieved and fairly good control performances are obtained.