This is the abstract of a talk prepared for the International interdisciplinary seminar on new robotics, evolution and embodied cognition (IISREEC).12th to 15th November 2002, Lisbon, Portugal
Abstract: Evolutionary Robotics allows the experimenter to try and re-create some cognitive phenomena of interest under controlled conditions, whilst minimising the preconceptions and prejudices that are put into the model. In this way Evolutionary Robotics is potentially a new scientific tool for studying the basics of cognition, perception and learning. A thought experiment can be tested on either real or simulated robots/agents, typically giving artificial evolution a free hand to design a real-time artificial neural network that generates behaviour selected to demonstrate the desired properties.
Some examples from within our research group at Sussex include: the origins of basic visual perception and object recognition; homeostasis and the ability of agents to overcome severe sensory distortion; the origins of communication; the circumstances under which learning becomes important; and the interactions between evolution and development. As these experiments are done in a very abstract setting, the implications can be very general and widespread.