View from the boundary

Re-implementing biological mechanisms on robots not only has technological application but can provide a unique perspective on the nature of sensory p rocessing in animals. To make a robot work, we need to understand the funct ion as part of an embodied, behaving system. I argue that this perspective suggests that the terms "representation" and "information processing" can b e misleading when we seek to understand how neurobiological mechanisms carr y out perceptual processes. This argument is presented here with reference to a robot model of cricket behavior, which has demonstrated competence com parable to that of the insect, but utilizes surprisingly simple central, pr ocessing. Instead it depends on sensory interfaces that are well matched to the task, and on the link between environment, action, and perception.