DGR-MAPPO: Enhancing Multi-Agent Cooperative Exploration in Unknown Environments with Distance-Aware Communication and Global Average Pooling

To address the challenges of local observation limitations, difficulties in global collaboration, and low exploration efficiency encountered in multi-agent collaborative exploration of unknown environments, this paper introduces a multi-agent proximal policy optimization (MAPPO) algorithm that incorporates distance awareness and gated recurrent unit optimization. The study constructs a composite observation space that integrates global maps with local perceptions. The stability of feature extraction is improved through the introduction of a global average pooling layer, while a distance-based dynamic communication mechanism is developed to facilitate map information sharing among neighboring agents. Experimental results demonstrate that this approach significantly reduces the path overlap rate in untrained complex maps, markedly enhances exploration coverage and training convergence speed, and confirms its effectiveness in improving the collaborative efficiency of multi-agents and their adaptability to varying environments.