This paper aims at the coordinated maneuvering task in urban battlefield,aiming at the environmental factors such as urban traffic obstacles, weather conditions, and damage degree,a multi-agent collaborative maneuvering behavior decision-making method based on the MADDPG algorithm is proposed. By defining a multi-agent state model and objective function, collaborative decision-making for time efficiency, path optimization, and collision avoidance is achieved. The experiment is based on open-source data from OpenStreetMap, simulating a scenario where three agents collaborate for reconnaissance. The results show that this method can rapidly improve the efficiency of agent collaboration and achieve stable performance after approximately 30,000 iterations. The agents can dynamically adjust their paths, avoid obstacles or choose the optimal route, demonstrating effective collaborative planning capabilities under non-strict obstacle avoidance. This method provides an effective solution for collaborative navigation problems in complex environments, significantly improving the collaboration among agents.