Efficient and accurate online computation of motor torques for robotic manipulators
Abstract
Realtime computation of the inverse dynamics of the robotic manipulators is required for ensuring robot control. This paper presents a modified NewtonEuler algorithm which makes use of symbolic programming for improved computational efficiency. Also, friction is incorporated in the dynamic model for more accurate prediction of torques. The algorithm is parallelized using a task streamlining approach, a systematic mapping scheme using layered task graphs to create the list schedule and a simplified binpacking heuristic algorithm to schedule the computations on a multiprocessor. The resulting computational load is only 12n+p flops (n=number of links in the manipulator), indicating a promise for application to precision robot control employing a high sampling rate. It has been shown that friction torques can be computed for greater accuracy of the dynamic model, modeling the joint friction using Coulomb's laws and the transmission friction, using the inputoutput function of the transmission system. The task streamlining approach was developed to decompose the inverse dynamic tasks into identical subtasks and to schedule in an efficient manner, resulting in a minimum processing time of 3.56 ms.
 Publication:

13th Canadian Congress of Applied Mechanics
 Pub Date:
 May 1991
 Bibcode:
 1991ccam.proc..654S
 Keywords:

 Dynamic Models;
 Inverse Kinematics;
 Manipulators;
 Newton Methods;
 OnLine Systems;
 Parallel Programming;
 Real Time Operation;
 Robot Arms;
 Robot Control;
 Symbolic Programming;
 Three Dimensional Motion;
 Torque;
 Algorithms;
 Euler Equations Of Motion;
 Friction;
 Heuristic Methods;
 Multiprocessing (Computers);
 Streamlining;
 Mechanical Engineering