Abstract: The generation of cooperation behaviors is the key point of multi-robot motion cooperation. In this article a new algorithm is presented for multi-robot motion cooperation under known circumstance, in which CMAC neural networks are properly employed to proximate the complex function maps from motion behaviors to circumstance states, and a novel genetic algorithm innovated from conventional GA-rational genetic algorithm (RGA) is used for generation and optimization of motion behaviors purpose. In such a way the robustness and completeness of the algorithm are improved, and the time complexity lowered.