Generalizable VLA Finetuning via Representation Anchoring and Language-Action Alignment

  • 类型:arxiv
  • 标识:2607.13429
  • 链接:https://arxiv.org/abs/2607.13429
  • 主分类:engineering
  • 形态:benchmark
  • 被引:1
  • 被引来源:Semantic Scholar
  • S2被引:1
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:Anchor-Align is proposed, which augments BC with two objectives: Vision-Language Anchoring distills layer-wise representations from a frozen VLM copy to prevent this drift, and Language-Action Alignment converts each action target into a discrete motion-direction label and jointly trains language and action prediction on the same robot observation.
  • OpenAlex ID:W7169081766
  • OpenAlex DOI:10.48550/arxiv.2607.13429
  • DOI:10.48550/arxiv.2607.13429
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2607.13429
  • OpenAlex更新:2026-08-24
  • 副分类:multimodal
  • 待LLM分类:否
  • 标题中文:通过表征锚定与语言-动作对齐的可泛化 VLA 微调
  • TLDR中文:本文提出 Anchor-Align,通过两个目标增强 BC:Vision-Language Anchoring 从冻结 VLM 副本中蒸馏逐层表示以防止该漂移;Language-Action Alignment 将每个动作目标转换为离散的运动方向标签,并在同一机器人观测上联合训练语言与动作预测。
  • 来源文件
  • /inbox/tom/_candidates/2026-07-23-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]