Attention has been increasingly focused on behaviorbased artificial intelligence (Al) due to its potential robustness and flexibility toward a dynamically changing world. This approach has yet unsolved problems: (1) how to construct an arbitration mechanism and (2) how to prepare competence modules (simple behavior/action). Biological information processing systems are interesting viewed from an engineering standpoint. Of these, we particularly have focused on the immune system, constructing a decentralized consensus-maker inspired by the immune network hypothesis. To solve the above problems in behavior-based Al, we apply our proposed method to behavior arbitration for an autonomous mobile robot in experiments using a real robot. We also study adaptation that automatically creates an artificial immune network using reinforcement signals.