Emergent Braitenberg-style Behaviours for Navigating the ViZDoom 'My Way Home' Labyrinth
Caleidgh Bayer, Robert J. Smith, Malcolm Iain Heywood · Proceedings of the Genetic and Evolutionary Computation Conference · 2025
The navigation of complex labyrinths under partially observable visual state is typically addressed using complex recurrent, convolutional learning architectures (i.e. deep reinforcement learning). Conversely, in this work, we show that navigation can be achieved through the emergent evolution of a simple Braitentberg-style vehicle. We demonstrate that the interaction between agent and labyrinth is sufficient to learn a complex navigation behaviour from simple heuristics. To do so, the approach of tangled program graphs is assumed in which programs cooperatively coevolve to develop a modular indexing scheme that employs < 2.5% of state space. We attribute this simplicity to several biases implicit in the representation, such as: (1) the use of pixel indexing as opposed to deploying a convolutional kernel or image processing operators, and; (2) extensive support for modularity in which behaviours are always decomposed into contexts and corresponding actions.