ACC99-IEEE1348 Control Comparisons for the Nonlinear Benchmark.pdf

ACC99-IEEE1348 Control Comparisons for the Nonlinear BenchmarkSystemJ. Willard Curtis Randal W. BeardElectrical and Computer Engineering DepartmentBrigham Young UniversityProvo, UT, 84602fwilly, beardg@AbstractThe experimental performance of recently proposedapproximation techniques for nonlinear optimal, andnonlinear H1 control compare favorably with otherpopular design techniques including LQR, linear H1,backstepping, and passivity based control. The per-formance of the aforementioned six control laws ona physical system similar to the Nonlinear Bench-mark problem proposed by Bernstein is compared andcontrasted. For the benchmark system, the succes-sive Galerkin approximation of the nonlinear H1 con-troller, appears to give superior results on a semi-global region of state space.1 IntroductionFor linear systems, LQR andH1 control are signi-cant design tools that have been successfully applied toa wide variety of practical systems. The counter partof these techniques for nonlinear systems are nonlinearoptimal control and nonlinear H1 control. Unfortu-nately, both of these techniques depend on the solu-tion to Hamilton-Jacobi partial dierential equationswhich are well known to be dicult to solve in gen-eral. Recently, the successive Galerkin approximation(SGA) technique has been introduced as a techniquefor approximating Hamilton-Jacobi-Bellman (HJB)and Hamilton-Jacobi-Isaacs (HJI) equations to an ar-bitrary level of accuracy on a region of state spacechosen by the designer [1, 2]. The approximation tech-nique has several advantages including explicit feed-back control laws. The SGA techniques has beensuccessfully applied to a number of practical systemsin both simulation [3, 4, 5] and in hardware [6, 7].The objective of this paper is to compare the ex-Corresponding Author. perimental performance of the SGA approximationof nonlinear optimal (SGA/HJB) and nonlinear H1(SGA/HJI) control laws to other popular design tech-niques for nonlinear systems. In particular,