Improved performance of naïve creature learning to cross a highway

Abstract

We present a model of simulated naïve creatures learning to cross a highway. The creatures are able to learn by "imitating what works and avoiding what does not work". Our simulations show that the creatures' population success in learning to cross a single-lane unidirectional highway is influenced by the conditions of the environment; i.e. by the car traffic density, presence or not of "erratic drivers", and by the selection of a crossing point location on the highway. The creatures' population success of crossing the highway can be improved by allowing the creatures to move randomly along the edge of the highway to find a safer location to cross from, and by preserving the knowledge base from environment with lower car traffic density to the one with higher car traffic density.