Sensorimotor transformations in the worlds of frogs and robots

Abstract

The paper develops a multilevel approach to the design and analysis of systems with “action-oriented perception”, situating various robot and animal “designs” in an evolutionary perspective. We present a set of biological design principles within a broader perspective that shows their relevance for robot design. We introduce schemas to provide a coarse-grain analysis of “cooperative computation” in the brains of animals and the “brains” of robots, starting with an analysis of approach, avoidance, detour behavior, and path planning in frogs. An explicit account of neural mechanism of avoidance behavior in the frog illustrates how schemas may be implemented in neural networks. The focus of the rest of the article is on the relation of instinctive to reflective behavior. We generalize an analysis of the interaction of perceptual schemas in the VISIONS system for computer vision to a view of the interaction of perceptual and motor schemas in distributed planning which, we argue, has great promise for integrating mechanisms for action and perception in both animal and robot. We conclude with general observations on the lessons on relating structure and function which can be carried from biology to technology.