Adaptive control of a dynamic system using genetic-based methods

The authors present genetic-based learning algorithms for automatically inducing control rules for a typical unstable, multioutput, dynamic system, namely, a simulated pole-cart system. They compare the performance of the genetic method with that of other learning algorithms for the same task. The experiments demonstrate that the results obtained with the genetic-based controller are comparable to those of existing methods. A further enhancement of genetic learning is possible by applying the structured genetic algorithm, which appears to offer improvements over the simple genetic algorithm in terms of robustness and speed of optimization.<<ETX>>