Automatic Evolution of Multimodal Behavior with Multi-Brain HyperNEAT

摘要: An important challenge in neuroevolution is to evolve multimodal behavior. Indirect network encodings can potentially answer this challenge. Yet practice, indirect do not yield effective controllers. This paper introduces novel extensions HyperNEAT, a popular encoding. A previous approach called situational policy geometry assumes that multiple brains benefit from being embedded within an explicit geometric space. However, HyperNEAT for evolving many without assuming relationships between them. The resulting Multi-Brain exploit human-specified task divisions, or automatically discover when should be used, and how use. Experiments show multi-brain approaches are more than extensions, relation each other superior.