Failure-Free Genetic Algorithm Optimization of a System Controller Using SAFE/LEARNING Controllers in Tandem

TL;DR

This paper introduces a failure-free genetic algorithm optimization method for system controllers by using SAFE and LEARNING controllers together, ensuring safe operation during the learning process, validated on an inverted pendulum system.

Contribution

It proposes a novel tandem SAFE/LEARNING controller approach to enable failure-free genetic algorithm optimization of system controllers.

Findings

Successfully applied to an unstable inverted pendulum system.

Ensures safety during the learning process.

Demonstrates effective system stabilization during optimization.

Abstract

The paper presents a method for failure free genetic algorithm optimization of a system controller. Genetic algorithms present a powerful tool that facilitates producing near-optimal system controllers. Applied to such methods of computational intelligence as neural networks or fuzzy logic, these methods are capable of combining the non-linear mapping capabilities of the latter with learning the system behavior directly, that is, without a prior model. At the same time, genetic algorithms routinely produce solutions that lead to the failure of the controlled system. Such solutions are generally unacceptable for applications where safe operation must be guaranteed. We present here a method of design, which allows failure-free application of genetic algorithms through utilization of SAFE and LEARNING controllers in tandem, where the SAFE controller recovers the system from dangerous states while the LEARNING controller learns its behavior. The method has been validated by applying it to an inherently unstable system of inverted pendulum.