In the standard approach to neuroevolution, the champion of the final generation is selected as the end result. However, it is possible that there is valuable information present in the population that is not captured by the champion. The standard approach ignores all such information. One possible solution to this problem is to combine multiple individuals from the final population into an ensemble. This approach has been successful in supervised classification tasks, and the goal of this project is to extend the approach to evolutionary reinforcement learning in control problems. The initial implementation of this approach uses NEAT (NeuroEvolution of Augmenting Topologies) to evolve both a population and a gating network that combines members of the population into an ensemble.
This approach has been tested on a challenging extension of the classic pole balancing problem in which the goal is to bring the end of the pole as close as possible to a moving particle, and the use of an ensemble improved performance over that of the champion alone.
Video of the Pole Chasing problem