A real-time computer architecture for inverse kinematics

Abstract

A special-purpose computer architecture has been developed for Inverse Kinematics so as to achieve real-tlme capabilities. The algorithm used is based on a modified predictor-corrector method for numerical integration and shows robustness even near points of singularity. The architecture is based on a 32-blt floating point unit that is equipped with data paths that facilitate computation of the most common matrix-vector operations used in robotics control. Special algorithms are used for the sine/ cosine and reciprocal, and these use a mix of computation and table lookup. The algorithm has been simulated on the proposed architecture and the results show its robustness and real-time capability. Two manipulator models are used, one of which does not have an analytical solution to its kinematics. The results indicate that its Inverse Kinematics solution may be obtained at more than 2000 points per second with an error within standard repeatability limits for industrial robots.