Dynamic fields endow behavior-based robots with representations

摘要: Abstract The behavior-based approach to autonomous robotics has been quite succesful at designing systems capable of simple reactive behaviors that are flexibly activated. Progress towards larger behavioral complexity and the equivalent cognitive abilities hampered, however, by very absence representations in this approach. Also, a need for theoretically analyzable general architecture system integration recognized. We propose an based on dynamic neural fields which aims deal with both these shortcomings. Sensory processing, representation, planning, control all addressed same terms defining adequate levels description. At each level field strong internal interactions is generating behavior or its representation. Levels mutually coupled specifying other representations. demonstrate solving standard problem target acquisition obstacle avoidance vehicle moving two dimensions. limited viewing angle visual sensor used exemplary fashion properties subsymbolic memory as implemented fields. show how representation alleviates spurious states potential approaches. we enhances system's capability escape from closed-in situations (e.g., canyon boxes), and, more generally, endows features.