Group animation has been a challenging research topic in the field of film production. In view of the fact that current digital emulation simulation of the movie battle scene is flawed with the problems of single motion mode, poor authenticity and heavy workload, this paper proposes a method of group simulation for film melee scenes based on the multiple autonomous agents. First, the area of operation is intelligently divided according to the scale and angle of the scene, and the corresponding local KD tree is established. Then, on the basis of visual perception, obstacle avoidance, evasion, and intelligent invoking of action data, the group roles in the melee scenes are intelligently controlled, achieving group behaviors such as group follow-up, grouping, the intelligent search target, and so on. The experimental results show that the proposed methodis feasible, and that it could simulate the multi-role intelligent movement of the complex film melee scenes well with high fidelity.