---
title: "意图蒸馏：提升视觉-语言-动作模型的语义目标理解"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.23478v1"
permalink: "https://feed.astralor.com/items/itm_6c9da22dfef8ac27b572eb97"
published: "2026-08-24T16:42:49.000Z"
score: 75
channel: "ai"
tags: ["GR00T N1.7","VLA 模型","π0.5","具身智能","意图蒸馏","机器人学习"]
---

# 意图蒸馏：提升视觉-语言-动作模型的语义目标理解

## AI 点评

> 这篇论文解决了 VLA 模型训练中一个容易被忽视的问题——行为克隆只学动作、不懂意图。INDI 的思路很聪明，直接用教师 VLM 把目标蒸馏进解码器，不需要额外的未来信息模块，落地成本低。对做机器人操作和具身智能的团队尤其有价值，特别是面对长时程任务时，它能明显缩小与人类示范之间的语义鸿沟。

## 速览

视觉-语言-动作模型通常通过行为克隆学习动作解码器，缺乏对行为背后语义目标的理解。本文提出意图蒸馏（INDI）方法，在训练时用冻结的教师 VLM 将演示片段的意图表征蒸馏到动作解码器，部署时从中间层恢复多模态意图表征，并用于组织动作预测。在 SimplerEnv-Bridge 上，INDI 将 GR00T-N1.7 成功率从 64.3% 提升至 84.7%；RoboCasa Kitchen 上从 64.1% 提升至 70.3%；真实世界任务平均成功率从 62.0% 提升至 68.7%，长时程任务提升达 12.0 个百分点。分析表明，恢复的潜在意图被解码器用于捕获行为目标与执行进度，显著提升模型泛化能力。

## 关键点

- 提出 INDI 方法，通过教师 VLM 将行为级别的意图蒸馏到动作解码器，显式建模行为语义目标。
- 在多个仿真与真实世界基准上显著提升 GR00T-N1.7 和 π0.5 的成功率，长时程任务增益更大。
- 实验证明解码器利用恢复的意图表征来捕获行为目标与执行进度，并依赖目标组织动作预测。

## 全文

Vision-Language-Action (VLA) models can turn multimodal context into robot actions, but their action decoders are still trained largely by behavior cloning. This supervises which motor command was demonstrated while leaving implicit the local objective served by the behavior under the instruction. Future-based supervision enriches action learning with frames, latent observations, trajectories, or motion representations, but these signals capture particular realizations of what may happen rather than the shared semantic objective of the forthcoming behavior. We propose Intention Distillation (INDI), which distills behavior-level intent into the action decoder. During training, a frozen teacher VLM interprets a demonstrated segment from the current observation, instruction, coarse action summary, and corresponding execution video. From its standard inputs, the deployed VLA recovers the resulting multimodal intent representation at an intermediate decoder layer and uses it to organize action prediction together with representations of how the behavior unfolds and what it achieves. On SimplerEnv-Bridge, INDI improves GR00T-N1.7 from 64.3% to 84.7%, and on RoboCasa Kitchen it improves the controlled GR00T-N1.7 baseline from 64.1% to 70.3%, with consistent gains on $π_{0.5}$ across both benchmarks. In real-world tasks, INDI improves average success from 62.0% to 68.7%, with gains of up to 12.0 pp on longer-horizon tasks. Further analyses show that the recovered latent is used by the decoder, captures behavior objective and execution progress, and organizes downstream predictions in an objective-dependent manner. These results show that action decoders benefit from explicitly modeling the semantic objective of the behavior they generate.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_6c9da22dfef8ac27b572eb97