A systolic array chip for robot inverse dynamics computation

To ensure smooth and accurate movement of a robot arm, the robot inverse dynamics problem must be solved at each servo sampling. The computation of this problem, however, is a mathematically intense task which degrades the sampling period of presentday robot control systems. In addition to the repetitive requirement for its evaluation, the linearly recursive and computer-bound properties of the robot inverse dynamics problem using the Newton-Euler (N-E) equations of motion suggest that it is amenable for direct mapping onto a fixed systolic array structure. This paper presents such an architecture and discusses its implementation in 1-micron CMOS technology, to compute the N-E algorithm for an n-link manipulator, within a period of 69+12n clock cycles. For a six-link robot manipulator operating at the maximum device frequency of 25 MHz, the total execution time is 5.64 μs. The die size of this robot controller chip is 530×485 square mils, and its estimated power dissipation at the specified frequency is 3.5 watts