Abstract

Animats-autonomous robots or simulations of animals-and the animat approach represent the most recent attempt to comprehend the capacity of animals for autonomous generation of adaptive, intelligent behavior in complex, changing environments. Motivated by perceived limitations in classical artificial intelligence (AI), the animat approach promulgates an alternative, bottom-up route to understanding intelligent behavior. Important tenets include: (1) that adaptive behavior is best understood by focusing on the interaction between a behaving individual and its environment, hence the interest in `embodied' physical robots `situated' in natural environments; (2) that specific abilities, `behaviors', are more natural units of analysis and design than general, information-processing functions and world models; and (3) that high-level behaviors will emerge as systems composed of simple behavioral competences become more complex. Thus, animat research often begins with low-level sensorimotor abilities and then moves up towards higher, cognitive functions. Both in analysis and in design, the animat approach borrows heavily from ethology, psychology, neurobiology and evolutionary biology, as well as from connectionism. For AI and robotics researchers, understanding the mechanisms behind adaptive behavior is secondary to creating them, but natural scientists can hope for tools and concepts to aid understanding of biological systems.