PRIME W0：面向具身操作的世界模型物理先验
PRIME W0: Physical Priors from World Models for Embodied Manipulation
Huabin Peng · Fagang Jin · Yancong Wei · Tengfei Ma · Chuhang Liu · Chao Yu · et al.
中文摘要
视觉-语言-动作（VLA）模型擅长遵循指令，但大多仍是前馈式：仅以当前观测为条件生成动作，丢弃了长时程时间上下文与 VLM 表征的层次结构。PRIME W0 是一个层次化条件流匹配（flow-matching）框架，通过三项协同设计弥补这一差距：(1) 将预训练 VLM 的逐层 KV 注入自适应 MoE 动作专家；(2) 通过异步世界模型子目标图像预测与 DINOv2 历史帧语义桥实现动态感知条件化；(3) 采用协方差结构化噪声的 Action Dynamic Manifold（ADM）流匹配，以尊重机器人运动学耦合。PRIME W0 分四阶段训练——VLM 预训练、DAT/NTP 视觉运动预训练、带 WM 注入的 DiT 流匹配解码器组装、端到端 SFT——在 LIBERO 与真实机器人基准上取得 state-of-the-art 结果，并通过系统性消融验证各架构组件。
关键要点
- 01现有 VLA 多为前馈式，丢弃长时程时序上下文与 VLM 层次结构信息
- 02逐层 KV 注入预训练 VLM 至自适应 MoE 动作专家，实现层次化条件
- 03异步世界模型预测子目标图像，并用 DINOv2 历史帧语义桥提供动态条件
- 04ADM 流匹配用协方差结构化噪声建模机器人运动学耦合
- 05四阶段训练后在 LIBERO 与真实机器人基准达 SOTA，消融验证各组件
解读
尚无解读。
原始英文摘要
arXiv:2610.39528v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models excel at instruction following, yet most remain feed-forward: they condition actions only on the current observation and discard both long-horizon temporal context and the hierarchical structure of VLM representations. We present PRIME W0, a hierarchical conditional flow-matching framework that closes this gap through three synergistic designs: (1)per-layer KV injection from a pretrained VLM into an Adaptive MoE action expert; (2)dynamics-aware conditioning via asynchronous world-model subgoal-image prediction and a DINOv2 historical-frame semantic bridge; and (3)Action Dynamic Manifold (ADM) flow matching with covariance-structured noise that respects robot kinematic coupling. PRIME W0 is trained in four stages---VLM pre-training, DAT/NTP visuomotor pre-training, DiT flow-matching decoder assembly with WM injection, and end-to-end SFT---and achieves state-of-the-art results on LIBERO and real-robot benchmarks, with systematic ablations validating each architectural component.