Robots are complex entities that can be modeled as multi-agent systems. The multi-agent paradigm provides an integrated intelligence framework such as a path planning agent that uses search techniques interacts with a fuzzy-based agent that moves the robot to a given location. Agent coordination is required to achieve the appropriate global behavior. When there is no central agent that coordinates the overall architecture, the intelligence required for social interaction should therefore be deployed at the agent level. In such a situation, individual intelligence (how to reach a goal) and social intelligence (how to collaborate or compete for resource with other agents) should be integrated at the agent level. In this paper we propose the use of module-based agents to achieve this integration. The whole multi-agent robot architecture, ARMADiCo, has been implemented with several module-based agents and tested on a Pioneer 2DX of ActivMedia. Some preliminary results are shown and discussed.