论文/arXiv 2610.13740·19 天前

SmolWAM：用于边缘设备机器人操作的潜变量世界-动作模型

SmolWAM通过重建Cosmos Policy推理接口，降低WAM边缘部署开销，并在400个LIBERO episodes上验证。

AI 整理

• 具身模型
• 端侧部署

原文链接：SmolWAM: A Latent World-Action Model for Robotic Manipulation on Edge Devices（arXiv 2610.13740 · 发布日期：2026-10-09）

SmolWAM: A Latent World-Action Model for Robotic Manipulation on Edge Devices

SmolWAM通过重建Cosmos Policy推理接口，降低WAM边缘部署开销，并在400个LIBERO episodes上验证。

核心要点

• 现有强WAM依赖视频扩散Transformer，计算开销过高，不适合资源受限边缘设备。
• SmolWAM保留Cosmos Policy预训练主干，改用单帧视觉潜变量、缓存文本特征等接口。
• 模型使用12/28个Transformer块和1,008个token，一步读出动作与未来视觉潜变量。
• 经3.6 GPU小时训练，在400个LIBERO episodes中成功率95.5%，且快于教师模型。

验证范围与局限：验证范围限于400个LIBERO episodes。

英文摘要 (Abstract)

World-action models (WAMs) predict a robot's actions together with the observations those actions will produce. However, the strongest models rely on video diffusion transformers that are far too computationally expensive for resource-constrained edge devices. We present SmolWAM, a latent WAM that we derive from the 2B-parameter Cosmos Policy by keeping its pretrained trunk and rebuilding its inference interface: single-frame visual latents, cached text features that preserve the trunk's attention sinks, 12 of 28 transformer blocks, and 1,008 tokens from which actions and future visual latents are read out in one rectified-flow step. Trained in 3.6 GPU-hours, SmolWAM succeeds in 95.5% of 400 LIBERO episodes at 57 ms and 2.3 TFLOP per 16-action chunk with 2.4 GiB of GPU memory, 11x faster than its teacher with 18.5x fewer FLOPs and 3.2x less memory, and its world head still forecasts scene motion.

作者：Timi Owolabi, Hikmah Olawore, Frances Adelakun, Ayotomiwa Oyewumi

来源

本页事实对应的公开页面，以原文为准。

1. arXiv 2610.13740：SmolWAM: A Latent World-Action Model for Robotic Manipulation on Edge Devices访问 2026-10-21