TravExplorer: Cross-Floor Embodied Exploration via Traversability-Aware 3-D Planning

TL;DR AI
2 min readKey summary
TravExplorer is a new embodied navigation system for zero-shot object search in unseen indoor buildings.
It combines semantic guidance, a unified 3-D map, traversable frontier extraction, active perception, and hierarchical planning.
The system can navigate floors, stairs, and landings, enabling cross-floor exploration without prior maps.
In simulation and real-world tests, including on a Unitree Go2 robot, it outperformed baseline ObjectNav methods.
