The implementation of complex behavior generation for arti®cial systems can be overcome by decomposing the global tasks into simpler, well-speci®ed behaviors which are easier to design and can be tuned independently of each other. Robot behavior can be implemented as a set of fuzzy rules which mimic expert knowledge in speci®c tasks in order to model expert knowledge. These behaviors are included in the lowest level of a hybrid deliberative±reactive architecture which is aimed at an ecient integration of planning and reactive control. In this work, we brie¯y present the architecture and attention is focused on the design, coordination and fusion of the elementary behaviors. The design is based on regulatory control using fuzzy logic control and the coordination is de®ned by fuzzy metarules which de®ne the context of applicability for each behavior. Regarding action fusion, two combination methods for fusing the preferences from each behavior are used in the experiments. In order to validate the system, several measures are also proposed, and thus the performance of the architecture and combination/arbitration algorithms have been demonstrated in both the simulated and the real world. The robot achieves every control objective and the trajectory is smooth in spite of the interaction between several behaviors, unexpected obstacles and the presence of noisy data. When the results of the experimentation from both methods are taken into account, the in¯uence of the combination method appears to be of prime importance when attempting to achieve the best trade-o among the preferences of every behavior.