This paper proposes an embedded intelligent robot model based on Multi-agent, and implements it on a robot dog platform with embedded, modular components. In order to make the software structure flexible and does not depend on specific hardware, we design and divide robot into few agents by the idea, which is “thinking and learning, perception and interaction, action and control”. What's more, the model takes account of both intelligence and response capability for emergency by deliberative module and reflex module. The mechanism of Case-based Reasoning is applied to realize the deliberation of robot dog. In the simulation and actual integration test, the robot dog is able to walk, avoid obstacles, perceive and process external signals such as voice input and infrared. The results proved that preliminary success was achieved.