Extending Behavior-Based Systems Capabilities Using An Abstract Behavior Representation

Behavior-based systems BBS have been e ective in a variety of applications, but due to their limited use of representation (sentence or logic-like structures) they have not been used much for more complex problems involving sequences of behaviors and they have been typically constructed by hand for each task. In this paper, we present an abstract behavior representation that allows for automatically specifying behavior networks thatbehavior representation that allows for automatically specifying behavior networks that encode complex behavioral sequences, based on a given set of underlying behaviors, and avoids customized behavior redesign while accommodating the speci cs of a new task. The representation, obtained by separating behaviors into two classes, abstract and primitive, allows BBS to generate and maintain complex plan-like strategies as well as switch them at run-time, without any need for behavior redesign and/or recompilation. To validate the described representation we have performed two object delivery tasks, involving behavior con icts and various initial conditions, using a Pioneer2 DX mobile robot.