Frontier-Based Exploration for Heterogeneous Multi-Agent Semantic Navigation

In visual semantic navigation, agents are tasked with locating and reaching multiple semantic targets in complex indoor environments using visual and semantic cues. Existing approaches typically adopt either a single agent or multiple homogeneous agents with identical capabilities. However, these configurations often suffer from limited perception and inefficient exploration. To address these challenges, We propose a heterogeneous multiagent visual semantic navigation framework that leverages the complementary perceptual capabilities of diverse agents. The system periodically fuses local observations to construct a unified global semantic map. A frontier-based exploration strategy, combined with centralized task allocation, enables efficient multi-target search. In complex and unseen indoor environments, the proposed method achieves higher task success rates and improved path efficiency compared to traditional single-agent baselines. The results highlight the advantages of heterogeneous collaboration in enhancing system performance and robustness under uncertainty.