TARIC: Memory-Augmented Traversability-Aware Outdoor VLN Under Interrupted Semantic Cues
Researchers have introduced TARIC, a comprehensive vision-language navigation (VLN) framework for outdoor environments that addresses interruptions caused by semantic cues over long distances. In these scenarios, valuable goal cues can become infrequent, blocked, or completely out of sight, leading agents to revert to backtracking, erratic movements, or aimless wandering. Existing memory-based approaches struggle with detours driven by traversability, as the remembered cue direction may not be practical, resulting in extended periods without cues and fading robot-centric cues. TARIC redefines traversability as essential for consistent goal-oriented navigation rather than just a local safety issue. It ensures stable, executable guidance during extended cue-free intervals, allowing for resilience against semantic-cue interruptions. The research can be found on arXiv with ID 2605.31121.
Key facts
- TARIC is a unified outdoor VLN framework.
- It addresses semantic-cue interruptions in long-range, open-world environments.
- Goal cues can become sparse, occluded, or leave the field of view.
- Agents often degrade into backtracking, oscillatory headings, or aimless exploration.
- Memory-based methods fail under traversability-driven detours.
- Traversability is treated as a stability condition for goal-directed guidance.
- The framework maintains traversability-consistent executable guidance during cue-free phases.
- The paper is on arXiv with ID 2605.31121.
Entities
Institutions
- arXiv