UniLAP：用于跨 embodiment 操作的统一潜在动作策略
UniLAP: A Unified Latent Action Policy for Cross-Embodiment Manipulation
AI总结：
UniLAP 是一种统一连续潜在动作策略，通过流匹配建模潜在空间，蒸馏 LAM 教师并解耦外观与运动线索，以 11 亿参数预训练后在多操作任务中优于现有基线。
摘要：
潜在动作模型（LAMs）可从无动作视频中捕获操作先验，但离散潜变量通常会造成粒度瓶颈，限制其表示细粒度动作的能力。近期的连续替代方案缓解了这一问题，但它们通常被用作世界建模的代理，且需要进一步的规划步骤来生成可执行动作。因此，我们引入 UniLAP，一种统一的连续潜在动作策略，通过流匹配对潜在空间进行建模，并在推理过程中绕过对未来帧的依赖。通过采用流自适应目标蒸馏 LAM 教师模型，UniLAP 继承了教师的动作表示，同时保留了可行行为的多模态分布。为了将潜在表示与任务无关的外观解耦，我们提取无标签运动目标，以从虚假视觉动态中分离出相关运动学线索。凭借已建立的与 embodiment 无关的流形，UniLAP 可轻松适配新平台。UniLAP 以紧凑的 11 亿参数规模在多源视频上进行预训练，在手部姿态预测、单臂及人形机器人操作和实际部署任务中，均优于最先进的潜在动作和视觉-语言-动作（VLA）基线。
英文摘要：
Latent Action Models (LAMs) can capture manipulation priors from action-free videos, but discrete latents usually impose a granularity bottleneck, limiting their ability to represent fine-grained action. Recent continuous alternatives mitigate this, but they are typically used as proxies for world modeling and require a further planning step to produce executable actions. We thus introduce UniLAP, a unified continuous latent-action policy that models the latent space via flow matching and bypasses reliance on future frames during inference. By distilling a LAM teacher with a flow-adaptive objective, UniLAP inherits the teacher's action representation while retaining the multimodal distribution of feasible behaviors. To decouple latent representations from task-irrelevant appearance, we extract a label-free motion target to isolate relevant kinematic cues from spurious visual dynamics. With the established embodiment-agnostic latent manifold, UniLAP can be easily adapted to novel platforms. Pretrained on multi-source videos at a compact 1.1B parameter scale, UniLAP outperforms state-of-the-art latent-action and Vision-Language-Action (VLA) baselines across hand pose forecasting, single-arm and humanoid manipulation, and real-world deployment.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
