To address the challenges faced by flipper tracked robots in achieving flipper autonomous control in a 3D environment, a flipper control method based on multi-agent reinforcement learning is proposed. Consider each flipper of the robot as an independent intelligent agent, design a reward function that balances chassis stability and flipper movements, and use multi-agent reinforcement learning to train the movements of each flipper; Deploy the proposed method in a 3D simulation environment based on Isaac Sim, and output the flipper angle by inputting local elevation maps and robot states to each agent. The experimental results show that this method can achieve autonomous control of the flipper in various terrains, and has significant improvement in robot autonomous obstacle crossing compared to single agent reinforcement learning.