Huang, Allen. “Assessing the feasibility of Novelty Search in the generalised pole balancing domain.”
URL1 URL2
The standard pole balancing problem, an effective benchmark for controllers in simulated environments is often ineffective for real-world scenarios. We implement a more generalised version of the pole balancing problem in the effort of creating controllers with real-world relevance. Additionally, all previous neuro-evolution approaches on the pole balancing problem were objective-based. As the size of the search space increases for the generalised version, we test a non objective-based neuro-evolution approach and compare its performance with the objective-based approach. The results of this thesis show that the non objective-based approach is not as effective at finding controllers for the generalised version of the pole balancing problem and the poor performance may be due to the lack of identifiable deception in the pole balancing domain.