Distributed behavior arbitration network: Autonomous behavior control architecture for humanoids

Abstract

AbstractA new concept of decentralized behavior control architecture for humanoids, termed the Distributed Behavior Arbitration Network, is proposed in this paper. First, policies for constructing the architecture are discussed and then the proposed architecture is designed according to these policies. In the proposed architecture, each competence of a humanoid is differentiated based on the behavior‐based framework and individual agents represent these differentiated competences. An agent consists of NetIBSI, which describes behavior patterns of the agent using state transition network, and the Conflict Resolution Strategy, which mediates with other agents. Autonomous distributed cooperation among agents becomes possible through these two components. The effect of the cooperation among multiple agents is to achieve various behavior patterns for humanoids. The proposed architecture was implemented in two types of humanoids constructed in our laboratory. It was able to realize fall avoidance behavior that required quick response and goal search behavior that required cooperation between two agents. The obtained results demonstrate that the proposed architecture is capable of generating complicated and intelligent behaviors for humanoids. © 2002 Wiley Periodicals, Inc. Syst Comp Jpn, 33(6): 32–43, 2002; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.1130