Abstract:With rapid advances in intelligent computing and communication technologies, achieving comprehensive collaborative situational awareness for intelligent sensor swarms in unknown environments has become feasible. However, obstacle occlusion can distort sensor perception utility models, thus limiting the swarm’s collaborative situational awareness capability. Therefore, this study proposes a new multi-agent collaborative decision-making method based on deep reinforcement learning, termed decentralized dynamic sensing optimization with deep Q-learning (DDSO-DQL) algorithm. The proposed approach constructs a distributed policy network to evaluate state-action values, designs a hierarchical exploration mechanism to improve environmental exploration efficiency, and incorporates a historical state backtracking strategy to enhance decision-making stability. This framework enables nodes to perform collaborative perception and decision-making autonomously, supporting dynamic situational awareness and global swarm optimization. Experimental results show that, in unknown environments, the proposed algorithm improves the average situational awareness rate by 8.7% compared to traditional methods, significantly enhancing collaborative network autonomy and dynamic sensing quality, and providing key technical support for intelligent perception and self-healing in intent-driven networks.