Applying the agent to path planning tasks and enabling cross-domain coordination have become a prominent research focus in recent years. However, traditional methods have limitations in image perception and data communication when conducting cross domain collaborative control. In response, this paper proposes a communication-optimized cross-domain coordination algorithm based on PPO. First, a monocular depth estimation algorithm integrating multi-scale feature is introduced, which combines information across scale and depth in images to enhance the model’s understanding of local details and global scenes and reduce communication burdens. Subsequently, a reinforcement learning reward function incorporating communication optimization strategies is designed to comprehensively balance communication data collection and obstacle avoidance capability. It meets the communication constraints and operational stability requirements. Finally, experimental results demonstrate that the proposed algorithm outperforms other methods in terms of obstacle avoidance efficiency and communication data collection.