LifelongCrossNav: Persistent 3D Semantic Memory for Cross-Floor Multi-Object Navigation
A novel framework named LifelongCrossNav has been developed to tackle the difficulties associated with sequential multi-object navigation in unfamiliar multi-floor indoor spaces. This system utilizes a shared sparse 3D semantic voxel memory that gradually gathers geometric structures, traversability states, and vision-language features, enabling future object-goal inquiries to access previously gathered scene data without the need to reconstruct the map. To facilitate ongoing searches across different floors, LifelongCrossNav integrates support-aware 3D traversability mapping, stair-specific perception, and direction-aware stair navigation. A cohesive navigation policy manages exploration on the same floor and stair traversal between floors. The framework is elaborated in a paper available on arXiv (2608.07079), showcasing its potential to enhance autonomous navigation in intricate indoor environments.
Key facts
- LifelongCrossNav is a framework for sequential multi-object ObjectNav in unknown multi-floor indoor environments.
- It maintains a shared sparse 3D semantic voxel memory that accumulates geometric structure, traversability states, and vision-language features.
- The memory allows subsequent object-goal queries to retrieve previously acquired scene information without rebuilding the map.
- It combines support-aware 3D traversability mapping, stair-specific perception, and direction-aware stair traversal.
- A unified navigation policy coordinates same-floor frontier exploration and cross-floor stair traversal.
- The paper is available on arXiv with identifier 2608.07079.
Entities
Institutions
- arXiv