VIRAL：面向人形机器人移动操作的大规模视觉仿真到现实迁移
VIRAL: Visual Sim-to-Real at Scale for Humanoid Loco-Manipulation
对做视觉的研究者，这篇给出了一个可复用的配方：用大规模平铺渲染 + 覆盖光照/材质/相机参数/图像质量/传感器延迟的视觉域随机化，配合真实到仿真的相机与灵巧手对齐，把纯 RGB 策略零样本迁移到真机；其中「算力规模（数十 GPU）是 sim-to-real 成败关键」这一结论值得在视觉策略训练中直接借鉴。
原文依据 VIRAL combines large-scale visual domain randomization--over lighting, materials, camera parameters, image quality, and sensor delays
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论人形机器人走向真实世界部署的一个关键障碍是缺乏自主的移动操作技能。我们提出 VIRAL，一个视觉仿真到现实框架，完全在仿真中学习人形机器人的移动操作，并将其零样本部署到真实硬件。VIRAL 采用教师-学生设计：一个基于完整状态的特权强化学习教师，使用增量动作空间与参考状态初始化来学习长时程移动操作。随后，一个基于视觉的学生策略通过带平铺渲染的大规模仿真从教师处蒸馏得到，并以在线 DAgger 与行为克隆的混合方式训练。我们发现算力规模至关重要：将仿真扩展到数十块 GPU（最多 64 块）可使教师与学生的训练都变得可靠，而低算力情形往往失败。为弥合仿真到现实的差距，VIRAL 将大规模视觉域随机化——覆盖光照、材质、相机参数、图像质量与传感器延迟——与灵巧手和相机的真实到仿真对齐相结合。部署在 Unitree G1 人形机器人上，所得的基于 RGB 的策略可连续执行移动操作多达 54 个循环，在无需任何真实世界微调的情况下泛化到多样的空间与外观变化，并接近专家级遥操作性能。大量消融实验剖析了使基于 RGB 的人形机器人移动操作在实践中奏效所需的关键设计选择。
英文原文摘要 1436 字符
A key barrier to the real-world deployment of humanoid robots is the lack of autonomous loco-manipulation skills. We introduce VIRAL, a visual sim-to-real framework that learns humanoid loco-manipulation entirely in simulation and deploys it zero-shot to real hardware. VIRAL follows a teacher-student design: a privileged RL teacher, operating on full state, learns long-horizon loco-manipulation using a delta action space and reference state initialization. A vision-based student policy is then distilled from the teacher via large-scale simulation with tiled rendering, trained with a mixture of online DAgger and behavior cloning. We find that compute scale is critical: scaling simulation to tens of GPUs (up to 64) makes both teacher and student training reliable, while low-compute regimes often fail. To bridge the sim-to-real gap, VIRAL combines large-scale visual domain randomization--over lighting, materials, camera parameters, image quality, and sensor delays--with real-to-sim alignment of the dexterous hand and cameras. Deployed on a Unitree G1 humanoid, the resulting RGB-based policy performs continuous loco-manipulation for up to 54 cycles, generalizing to diverse spatial and appearance variations without any real-world fine-tuning, and approaching expert-level teleoperation performance. Extensive ablations dissect the key design choices required to make RGB-based humanoid loco-manipulation work in practice.