Comparative Implementation Studies of Five Adaptive Control Algorithms on a Class 1 Manipulator

Abstract

The paper compares the results of implementing five well known adaptive algorithms on the motion control of a planar revolute axis manipulator. The main emphasis is on the experimental results, and on the relative efficacy of each algorithm. However, we also draw comparisons of the structure of the algorithms, and the corresponding implications for discrete-time implementations. The main conclusion is that the decentralized version of the minimal control synthesis (MCS) algorithm of Stoten and Benchoubane (1, 2) resulted in excellent model-following, with relative noise insensitivity and minimum flops per sampling interval. MCS is classed as a direct adaptive controller (3). The two other direct adaptive controllers of Seraji (4) and Oh and Jamshidi (5) did not perform as well as MCS; model-following was not so close, noise was more of a problem, and the implemented forms were more complicated. The indirect adaptive controllers of Craig (6) and Stotine and Li (7) produced virtually identical results; the model-following was excellent, but noise and parameter convergence problems were evident. The indirect adaptive controllers also suffered from the necessity for plant parameter estimation.