DAgger Diffusion Navigation: DAgger Boosted Diffusion Policy for Vision-Language Navigation

2025-08-12

AI Summary

To address the global suboptimality and performance bottlenecks inherent in two-stage waypoint planning for Vision-and-Language Navigation with Continuous Environments (VLN-CE), this paper proposes the first end-to-end diffusion policy model that jointly models waypoint generation and path execution. We innovatively introduce conditional diffusion models into VLN, enabling direct learning of the multimodal (vision-language-action) joint distribution in continuous action space. Our approach integrates DAgger-based online expert demonstration, trajectory augmentation, and policy fine-tuning to enhance robustness and behavioral diversity in long-horizon navigation. On the R2R and CVD benchmarks, our method significantly outperforms existing two-stage approaches—even without relying on external waypoint predictors—demonstrating superior effectiveness and generalization in complex instruction grounding and continuous control. This work establishes diffusion policies as a viable and powerful paradigm for end-to-end VLN-CE.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Planning, Routing, and Scheduling: Planning with Language Models
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Vision-Language Navigation in Continuous Environments (VLN-CE) requires agents to follow natural language instructions through free-form 3D spaces. Existing VLN-CE approaches typically use a two-stage waypoint planning framework, where a high-level waypoint predictor generates the navigable waypoints, and then a navigation planner suggests the intermediate goals in the high-level action space. However, this two-stage decomposition framework suffers from: (1) global sub-optimization due to the proxy objective in each stage, and (2) a performance bottleneck caused by the strong reliance on the quality of the first-stage predicted waypoints. To address these limitations, we propose DAgger Diffusion Navigation (DifNav), an end-to-end optimized VLN-CE policy that unifies the traditional two stages, i.e. waypoint generation and planning, into a single diffusion policy. Notably, DifNav employs a conditional diffusion policy to directly model multi-modal action distributions over future actions in continuous navigation space, eliminating the need for a waypoint predictor while enabling the agent to capture multiple possible instruction-following behaviors. To address the issues of compounding error in imitation learning and enhance spatial reasoning in long-horizon navigation tasks, we employ DAgger for online policy training and expert trajectory augmentation, and use the aggregated data to further fine-tune the policy. This approach significantly improves the policy's robustness and its ability to recover from error states. Extensive experiments on benchmark datasets demonstrate that, even without a waypoint predictor, the proposed method substantially outperforms previous state-of-the-art two-stage waypoint-based models in terms of navigation performance. Our code is available at: https://github.com/Tokishx/DifNav.

Problem

• Overcoming sub-optimization in two-stage VLN-CE navigation frameworks
• Eliminating reliance on waypoint predictor performance bottlenecks
• Improving robustness in long-horizon instruction-following navigation tasks

Innovation

• End-to-end diffusion policy for VLN-CE
• DAgger-enhanced online training for robustness
• Multi-modal action modeling in continuous space