Computing robot Jacobians on meshes with multiple buses

摘要

This paper is a theoretical evaluation of the viability of computing robot Jacobians algorithms on parallel platforms. These algorithms are based on some popular sequential algorithms and the work here will outline the suitability of these algorithms for parallelization. The overall aim is to compute the Jacobian at high speeds, in situ, to enable the timely control of robot manipulators. The jacobian is used to map information from the robot world (joint or robot space) to the human world (Cartesian space), as well as calculating torques and forces in the joints induced by an external end-of-arm force. Many algorithms exist for calculating the Jacobian. In this work, three algorithms are mapped onto a powerful parallel architecture. Two techniques have been used in mapping the algorithms onto the mesh with multiple buses: recursive doubling and pipelining. All three algorithms were mapped using these techniques, although some algorithms are better suited to a particular technique. The parallel algorithms are compared against each other, and against the sequential versions.