In view of the traditional behavior fusionspsila insufficiency in behavior interaction, a layered behavior fusion method is presented. Behavior references are environmental information and robotpsilas moving direction and velocity, and behavior outputs are robotpsilas moving direction and velocity. Behavior is arranged in different layer according to its objectpsilas relation, lower behavior takes upper behaviorpsilas output as its references, the lowest behavior output the finally fusion results. Finally taking robot motion planning in unknown dynamic environment as research background, realized the layered behavior fusion architecture. The validity of the method is proved by simulation experiment.