# Embodied-Navigator：点选、思考、记忆与对齐以实现高效导航

> 原文链接：https://robotics.hellobike.com/papers/2026-08-18-2608-17512
> 频道：papers | 更新日期：2026-08-18 | 标签：arxiv, cs.RO
> 核心摘要：TAMP-Nav通过2D像素提示、选择性推理记忆和两级对齐提升具身导航，并在R2R-CE等实验中验证。

# Embodied-Navigator：点选、思考、记忆与对齐以实现高效导航

::: v-pre
> 原文链接：[Embodied-Navigator: Point, Think, Memorize, and Align for Efficient Navigation](https://arxiv.org/abs/2608.17512)（arXiv 2608.17512 · 发布日期：2026-08-18）

**Embodied-Navigator: Point, Think, Memorize, and Align for Efficient Navigation**

TAMP-Nav通过2D像素提示、选择性推理记忆和两级对齐提升具身导航，并在R2R-CE等实验中验证。

## 核心要点

- 现有方法将VLM强置于不自然动作空间，且推理调度僵硬、记忆管理低效。
- Pixel-to-3D动作表述让VLM选择2D像素，再投影到3D坐标供低层SLAM控制。
- 选择性触发Chain-of-Thought，并仅在关键节点保留高保真锚点轨迹记忆。
- 两级对齐用GRPO叠加结果奖励与过程奖励，在R2R-CE达66.2% SR且仅需90k训练轨迹。

**验证范围与局限**：摘要仅明确给出R2R-CE上的结果，未说明是否进行真机验证。

## 英文摘要 (Abstract)

Although Large Vision-Language Models (VLMs) have significantly advanced embodied navigation, their direct deployment remains challenging, as existing methods often force VLMs into unnatural action spaces that misalign with their 2D pre-training priors, compounded by rigid reasoning schedules and inefficient memory management. To overcome these limitations, we propose TAMP-Nav, a unified framework for efficient embodied navigation. First, we introduce a Pixel-to-3D Action Formulation (Point) that reformulates navigation into 2D visual prompting. Specifically, the VLM merely selects 2D pixels, which are then projected into 3D coordinates for a low-level SLAM controller. This design naturally aligns embodied execution with the VLM's inherent 2D visual capabilities. Second, we propose an integrated Selective Reasoning and Anchor-Trajectory Memory mechanism (Think and Memorize), which dynamically triggers Chain-of-Thought and retains high-fidelity memory only at critical nodes, compressing redundant trajectories into lightweight Space-Time Indicators, thereby preserving critical historical information and enhancing spatio-temporal perception. Finally, we design an efficient Two-Level Alignment Paradigm (Align) via Group Relative Policy Optimization (GRPO). By superimposing global outcome rewards with fine-grained process rewards, this dense supervision tightly aligns the agent's cognitive planning with physical environmental feedback, endowing the model with adaptive reasoning capabilities. Experiments demonstrate that TAMP-Nav achieves state-of-the-art performance (e.g., 66.2% SR on R2R-CE) with high runtime and sample efficiency (requiring only 90k training trajectories).

作者：Hongyan Feng, Sunlai Chen, Xuanyu Liu, Miao Pan, Yangfan Xie, Yuxiang Cui, Zhongxiang Zhou, Rong Xiong, Wenqi Zhang, Jianwei Yin
:::

## 原始参考来源 (Sources)

- [arXiv 2608.17512：Embodied-Navigator: Point, Think, Memorize, and Align for Efficient Navigation](https://arxiv.org/abs/2608.17512)