AAS 99-154

Limit Cycles and Optimized Error Levels in Automated Tuning of Learning Control Gains

S-L. Wirkander*, RW. Longman**

*National Defense Research Establishment, Stockholm, Sweden, **Columbia University, New York


Iterative learning control develops control laws that learn from previous experience executing a specific command in order to decrease the tracking error as the command is repeated. In previous work, learning control laws were developed that are similar to classical, P, PD, or PID control in the sense that one needs to tune only a small number of parameters. And, in learning control this results in a very substantial improvement in system performance with very little effort. The process of adjusting these parameters can be automated, creating self-tuning learning control, analogous to self-tuning regulators. This paper reports results of using a self-tuning scheme, where the learning process converges to a limit cycle behavior. This behavior is studied in detail, showing that limit cycles can very often produce improved tracking accuracy compared to learning control laws using fixed parameter choices.