Depth-Wise Probing and Pruning of the Planning Token in a Driving Vision-Language-Action Model

📄 arXiv: 2608.07361v1 📥 PDF

作者: Harisankar Babu, Benjamin Coors, Christopher Lang, Hendrik Berkemeyer, Tamim Asfour, Simon Foell

分类: cs.RO, cs.CV

发布日期: 2026-08-07

备注: Accepted at the 6th DriveX Workshop (Foundation Models for Autonomous Driving), ECCV 2026. 14 pages, 8 figures, 4 tables

💡 一句话要点

提出深度探测与剪枝方法以优化驾驶VLA模型的规划令牌

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 视觉-语言-动作 自动驾驶 模型剪枝 解码速度提升 深度学习

📋 核心要点

  1. 现有的VLA模型在驾驶决策中依赖深度语言模型，但不清楚动作所需的深度程度，存在效率低下的问题。
  2. 本文提出了一种通过探测和剪枝规划令牌的方法，旨在优化解码过程，提高模型的执行效率。
  3. 实验结果显示，剪枝后模型在保持较低误差的同时，实现了1.33倍的解码速度提升，验证了方法的有效性。

📝 摘要（中文）

本文研究了视觉-语言-动作（VLA）模型在驾驶决策中的应用，探讨了规划令牌的深度对动作需求的影响。通过对32个解码层的规划令牌进行探测，发现语义意图在早期层次上具有较高的线性可解性，而与原生规划器的兼容性则随着层次的加深逐步提高。研究表明，前几层的学习输出能够弥补这一差距，从而实现了在约5%的相对误差增加下，剪除8个解码层并获得1.33倍的解码速度提升。该研究结果限于评估的ORION检查点和Bench2Drive设置。

🔬 方法详解

问题定义：本文旨在解决驾驶VLA模型中规划令牌的深度对动作需求的影响，现有方法在效率和性能之间存在权衡。

核心思路：通过探测每个解码层的规划令牌，评估其对导航命令的线性可解性和与原生规划器的兼容性，从而识别可剪枝的层。

技术框架：整体架构包括32个解码层，通过对每层的输出进行分析，测量语义意图的可解性和与规划器的兼容性，最终确定可剪枝的层。

关键创新：提出了基于角度偏差对解码层进行排名的方法，允许在保持性能的同时剪除多层，显著提升解码速度。

关键设计：采用了线性可解性和兼容性作为评估指标，设计了相应的损失函数和参数设置，以确保剪枝后的模型性能不受显著影响。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，经过剪枝后，模型在相对开放环路误差增加约5%的情况下，成功去除了8个解码层，并实现了1.33倍的解码速度提升，验证了方法的有效性和实用性。

🎯 应用场景

该研究的潜在应用领域包括自动驾驶、智能交通系统和机器人导航等。通过优化VLA模型的解码过程，可以提高决策效率，降低计算资源消耗，从而推动智能驾驶技术的实际应用和发展。

📄 摘要（原文）

Vision-language-action (VLA) models route driving decisions through a deep language model, but it is unclear how much of that depth the action itself requires. We study a representative driving VLA whose entire plan is carried by a single planning token that a generative planner decodes into a trajectory. Borrowing the planner as a trajectory-space logit lens, we decode the planning token from every one of the 32 decoder layers and measure two signals: the linear decodability of the navigation command and trajectory compatibility with the frozen native planner. Our diagnostic shows that semantic intent is linearly decodable early: command-probe accuracy reaches 97.7\% after the first decoder layer, compared with 16.7\% chance. In contrast, compatibility with the frozen native planner improves gradually across depth, with open-loop Avg-L2 reaching its minimum of 2.11\,m only at the final layer. Learned readouts from the first layer recover much of this gap, indicating that planning information is already present early but is not yet represented in the format expected by the deployed planner. Ranking decoder layers by the angular deviation they induce in the planning token permits removal of 8 of 32 layers within an approximately 5\% relative open-loop error increase and yields a measured 1.33$\times$ decoder speedup. At the evaluated sample size, no family-specific degradation is statistically resolved. These findings are limited to the evaluated ORION checkpoint and Bench2Drive setup.