CoFL-S: Spatially Queryable Sector Flow Fields for Local Language-Conditioned Navigation

2026-07-02

AI Summary

This work addresses the limitation of existing vision-and-language navigation methods, which predominantly focus on high-level instruction understanding while neglecting continuous, fine-grained action generation. The authors propose CoFL-S, a novel framework that introduces, for the first time, a language-conditioned local sector flow field to predict dense motion directions within the robot’s visible region. By rolling this flow field over time, the method generates smooth trajectories, enabling fine-grained mapping from language instructions to motor actions. CoFL-S leverages frame-aligned sub-instructions and flow-field supervision signals, incorporates a continuous-time benchmark to decouple instruction decomposition from low-level control, and is evaluated in closed-loop settings under a unified velocity controller. Experiments demonstrate that CoFL-S significantly outperforms action-token and chunk-based baselines on the Habitat continuous-time benchmark and exhibits superior zero-shot generalization when deployed in real-world environments.

Abstract

Vision-Language Navigation has increasingly emphasized high-level instruction reasoning, memory, global map construction, and instruction decomposition, while the low-level action representation remains comparatively underexplored. We propose CoFL-S, a low-level vision-language-action framework that predicts a language-conditioned flow field over the robot's local visible sector and generates continuous trajectories by rolling out the predicted field. To train this low-level representation, we convert each VLN-CE episode, originally a whole-episode instruction paired with an action sequence, into frame-level local supervision with aligned sub-instructions and matched action, trajectory, and dense flow-field targets. For evaluation, we introduce a continuous-time Habitat benchmark that isolates low-level action interfaces from instruction decomposition and executes all methods through a shared velocity-command controller, enabling decomposition-independent closed-loop comparison across different planner frequencies rather than fixed discrete forward-and-turn transitions in VLN-CE. Under matched encoders and training settings, CoFL-S consistently outperforms action-token and action-chunk baselines across planner frequencies in the continuous-time Habitat benchmark, and zero-shot real-world closed-loop deployment further shows its advantage over both baselines beyond simulation.

Problem

• Vision-Language Navigation
• low-level action representation
• flow fields
• continuous trajectory
• language-conditioned navigation

Innovation

• flow fields
• continuous trajectory
• language-conditioned navigation
• low-level action representation
• continuous-time benchmark