This is a demo of robot foraging using Neuroevolution (specifically NEAT) with and without online Egalitarian Social Learning (ESL). The agents begin with randomly initialized networks, resulting in poor starting behaviors. Without ESL, they have fixed policies during each generation and do not improve during the episode. However, ESL agents are able to learn from observations of others in the population and quickly improve their policy. Thus, ESL agents begin to exhibit good behaviors before even the first generation has completed.
NEAT + ESL