BGC-Nav: Cross-Altitude Multi-UAV Collaborative Navigation via Belief-Guided Coupling

Key points

• Despite recent advances in Aerial Vision-and-Language Navigation (VLN), single Unmanned Aerial Vehicle (UAV) search-and-rescue in outdoor scenarios remains challenging due to partial observability, making cross-altitude collaboration a promising direction.1
• However, existing collaborative navigation methods rely on precise target localization and rigid goal-dispatching paradigms, ignoring high-altitude observation uncertainty and suppressing low-altitude exploration autonomy.1
• Specifically, the high-altitude UAV maintains a Global Belief Memory encoding task-conditioned priors, which the low-altitude agent incorporates into decision-making via Local Exhaustion Memory (LEM), Local Attraction Augmentation (LAA), and Directional Belief Prior (DBP) as an advisory bias rather than a rigid command.1
• Extensive experiments on ESARBench show that BGC-Nav consistently outperforms state-of-the-art single-agent and collaborative methods.1

Sources (1)

• [1]BGC-Nav: Cross-Altitude Multi-UAV Collaborative Navigation via Belief-Guided Coupling
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 10, 08:19 AM

  To address these issues, we introduce BGC-Nav, a training-free cross-altitude collaboration framework that enables robust navigation via Belief-Guided Coupling (BGC).

  Despite recent advances in Aerial Vision-and-Language Navigation (VLN), single Unmanned Aerial Vehicle (UAV) search-and-rescue in outdoor scenarios remains challenging due to partial observability, making cross-altitude collaboration a promising direction.

Extractive summary: sentences quoted from the sources.