ReflectVLN: Training Vision-Language Navigation Agents with Reflective Reasoning

2026-07-14

AI Summary

This work addresses the limitations of existing vision-and-language navigation (VLN) methods, which lack explicit closed-loop mechanisms to track semantic progress and recover from long-horizon errors. To overcome this, the authors propose ReflectVLN, a framework featuring a bidirectional interaction architecture between an intention agent and an execution agent. The intention agent decomposes tasks into subgoals and generates corrective plans, while the execution agent translates these into short-horizon actions and monitors navigational progress. Through closed-loop feedback, the two agents dynamically adjust and recover from deviations. The framework integrates an Action Chain-of-Thought training strategy and a path-conditioned dual-query mechanism, achieving significant improvements in success rate and path efficiency on standard VLN benchmarks. It also reduces training cost and the frequency of high-level intention invocations, while offering interpretable intermediate decision processes.

Abstract

Existing vision-language navigation methods often couple a VLM with waypoint decoders to produce multi-step action plans, but they typically lack an explicit closed-loop mechanism for tracking semantic progress, diagnosing execution failures, and recovering from error accumulation in long-horizon navigation. To address this gap, we propose ReflectVLN, an agentic VLN framework that organizes decision-making through bidirectionally interactive intention and execution agents. The intention agent performs subtask decomposition and reflection, generating executable subtask descriptions as corrective plans. Conditioned on these descriptions, the execution agent grounds them into short-horizon actions under current observations while monitoring sub-goal progress and detecting off-track behavior. Crucially, ReflectVLN enables closed-loop bidirectional communication: the execution agent emits progress and deviation signals to trigger reflection and subtask updates on demand, and the intention agent returns structured guidance that reconditions subsequent actions for recovery. To encourage temporally coherent decisions with interpretable intermediate rationales, we introduce Action Chain-of-Thought (Action-CoT), a path-conditioned dual-query training scheme for action generation. Experiments on standard VLN benchmarks show that ReflectVLN improves success rates and path efficiency under a constrained data budget, with favorable training cost and fewer high-level intention calls at inference time, while providing interpretable intermediate decisions for analysis and collaboration. Code is available at: https://github.com/AIprogrammer/ReflectVLN

Problem

• vision-language navigation
• closed-loop mechanism
• error accumulation
• semantic progress tracking
• execution failure diagnosis

Innovation

• Reflective Reasoning
• Vision-Language Navigation
• Closed-loop Planning
• Action Chain-of-Thought
• Agentic Framework