In order to meet the operational requirements of the increasingly complex battlefield environment, the planning method of transforming the UAV combat mode from single aircraft to group combat is proposed. In order to improve the efficiency of task decision-making before group action and reduce the potential risk of UAV between waypoints, a task decision-making method based on multi-agent reinforcement learning is introduced to improve the efficiency of autonomous decision-making, and the command generation in dynamic environment is realized; At the same time, aiming at the potential threats in the battlefield, a risk-Euclidean distance trensform (Risk-EDT) is designed, and then a risk avoidance method based on B-spline curve is proposed to avoid the potential threats between waypoints. Through the simulation experiment and physical experiment, the results show that the method has a high degree of applicability in the UAV cluster combat mission scene, can ensure the success rate of the mission, and improve the combat efficiency of the cluster.