发表机构
T-Stone Robotics Institute, The Chinese University of Hong Kong（香港中文大学T-Stone机器人研究所）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文提出Dense LSS方法，在训练阶段将VLA模型的动作标记与各操控阶段的物理推理依据对齐，推理阶段无额外开销，该方法在分布内任务及未见任务迁移上性能最优，可提升骨干网络的阶段可分性。
AI 中文摘要
视觉-语言-动作（Vision-language-action, VLA）模型通过模仿学习进行训练，能够掌握应采取的动作，但无法解释原因；引入因果推理可提升操控性能，不过现有方法在推理阶段需付出代价——每一步都要生成推理标记或展开预测的未来状态，这种代价会在长时序任务中不断累积。本文探究能否将因果推理的收益转移至训练阶段，而在部署前丢弃相关开销。我们提出潜在语义支架（Latent Semantic Scaffolding, LSS），这是一种在人类演示预训练阶段应用的辅助损失，通过小型投影头将VLA的动作标记表示与物理推理依据的文本嵌入对齐。推理阶段会丢弃该投影头，基础策略保持不变，无额外开销。核心发现在于对齐粒度：将每个动作标记与其自身操控阶段的推理依据对齐（Dense LSS，密集潜在语义支架），而非与单一的聚合回合级嵌入对齐（Pooled LSS，聚合潜在语义支架），能生成泛化性显著更优的表示。Dense LSS在分布内任务的成功率和对对齐阶段未见过任务的迁移性能均最优，而聚合对齐会过度适配训练任务。表征探测显示，Dense LSS使骨干网络的每阶段可分性提升约一倍，证明阶段局部对齐是其有效机制。
英文摘要
Vision-language-action (VLA) models are trained by imitation and capture what action to take but not why; adding causal reasoning improves manipulation, but current methods pay for it at inference time - generating reasoning tokens or rolling out predicted future states at every step, a cost that compounds over long horizons. We ask whether this benefit can instead be captured during training and discarded before deployment. We introduce Latent Semantic Scaffolding (LSS), an auxiliary loss applied during human-demonstration pretraining that aligns a VLA's action-token representations to text embeddings of physical-reasoning rationales through a small projection head. The head is dropped at inference, leaving the unmodified base policy with zero added cost. Our central finding concerns alignment granularity: aligning each action token to the rationale of its own manipulation phase (Dense LSS) rather than to a single pooled episode-level embedding (Pooled LSS) yields representations that transfer markedly better to held-out tasks. Dense LSS attains both the best in-distribution success and the best transfer to tasks unseen during alignment, whereas pooled alignment over-specializes to the training task. A representational probe shows Dense LSS induces roughly twice the per-phase separability in the backbone, supporting that phase-local alignment is the operative mechanism.
