GreenWAM: A World Foundation Model for Robotics
Overview
This paper introduces GreenWAM, a world foundation model for robotics that serves both world modeling and policy generation roles. It achieves top-tier performance on WorldArena 1.0 and ranks among the top entries in RoboTwin 2.0. The model is adapted for real-world humanoid robots and distilled into a single-step policy for real-time control, with controlled experiments highlighting the benefits of robot specialization and maintaining world-model modes during policy adaptation.
From arXiv Robotics — research abstracts
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · training_hours: 1674.7 hours · Basis not reported
“机器人数据 首先，通过数据注册确定每个数据源的摄像头角色、状态布局和机器人属性；随后，将数据切分为 原子样本 ，得到纳入语料的 144 万个完整子任务原子样本，总时长为 1,688.5 小时；经过质量筛选后，保留 143 万个原子样本，总时长为 1,674.7 小时；最后，通过生成描述文本和推导指令，为 119 万个原子样本生成了符合要求的指令。”
Exact source · revision 3Source owner not reported
Reported quantity · atomic_samples: 1190000 other · trajectory count
“机器人数据 首先，通过数据注册确定每个数据源的摄像头角色、状态布局和机器人属性；随后，将数据切分为 原子样本 ，得到纳入语料的 144 万个完整子任务原子样本，总时长为 1,688.5 小时；经过质量筛选后，保留 143 万个原子样本，总时长为 1,674.7 小时；最后，通过生成描述文本和推导指令，为 119 万个原子样本生成了符合要求的指令。”
Exact source · revision 3Source owner not reported
Reported quantity · total_hours: 1688.5 hours · Basis not reported
“机器人数据 首先，通过数据注册确定每个数据源的摄像头角色、状态布局和机器人属性；随后，将数据切分为 原子样本 ，得到纳入语料的 144 万个完整子任务原子样本，总时长为 1,688.5 小时；经过质量筛选后，保留 143 万个原子样本，总时长为 1,674.7 小时；最后，通过生成描述文本和推导指令，为 119 万个原子样本生成了符合要求的指令。”
Exact source · revision 3Source owner not reported
Reported quantity · video_samples: 1440000 other · trajectory count
“机器人数据 首先，通过数据注册确定每个数据源的摄像头角色、状态布局和机器人属性；随后，将数据切分为 原子样本 ，得到纳入语料的 144 万个完整子任务原子样本，总时长为 1,688.5 小时；经过质量筛选后，保留 143 万个原子样本，总时长为 1,674.7 小时；最后，通过生成描述文本和推导指令，为 119 万个原子样本生成了符合要求的指令。”
Exact source · revision 3Source owner not reported
Reported quantity · filtered_samples: 1430000 other · trajectory count
“机器人数据 首先，通过数据注册确定每个数据源的摄像头角色、状态布局和机器人属性；随后，将数据切分为 原子样本 ，得到纳入语料的 144 万个完整子任务原子样本，总时长为 1,688.5 小时；经过质量筛选后，保留 143 万个原子样本，总时长为 1,674.7 小时；最后，通过生成描述文本和推导指令，为 119 万个原子样本生成了符合要求的指令。”
Exact source · revision 3Source owner not reported
Reported quantity · instruction_samples: 1190000 other · trajectory count
“机器人数据 首先，通过数据注册确定每个数据源的摄像头角色、状态布局和机器人属性；随后，将数据切分为 原子样本 ，得到纳入语料的 144 万个完整子任务原子样本，总时长为 1,688.5 小时；经过质量筛选后，保留 143 万个原子样本，总时长为 1,674.7 小时；最后，通过生成描述文本和推导指令，为 119 万个原子样本生成了符合要求的指令。”
Exact source · revision 3Source owner not reported
Artifact availability · paper: available
“研究亮点： * 研究发现针对特定机器人进行专门化训练，能够同时提升世界预测与动作控制能力 * 研究发现在适配阶段开始时就将动作专家分离出来，有助于提升模型表现 * 研究发现在策略适配过程中保留世界模型的不同工作模式，有助于维持模型应对场景变化时的鲁棒性 论文地址： https://hyper.ai/papers/2610.18774 数据集：将 20 类机器人纳入统一动作空间 不同机器人数据集在帧率、摄像头布局、状态语义和任务边界等方面存在差异，双臂机器人、单臂机械臂和类人机器人即使执行相似任务，其原生动作表示也可能完全不同。”
Exact source · revision 3Source owner not reported
Evidence dependency changes (5)
- 2026-10-30T23:05:27.686Z · evidence updated · source revision 1. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T23:05:11.088Z · evidence updated · source revision 1. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T22:28:59.190Z · evidence updated · source revision 3. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T22:28:39.311Z · evidence updated · source revision 3. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-15T20:13:27.332Z · evidence updated · source revision 1. Evidence extraction was updated; current source attributions are shown above.
Report timeline
Follow attributed reports and material updates.
- Chinese robotics — Dataset and collection discoverySignalGreenWAM: A World Foundation Model for Robotics
A research paper introduces GreenWAM, a world foundation model for robotics that unifies world modeling and action control. The model is trained on a unified action space covering 20 robot types and demonstrates strong performance in trajectory prediction, depth estimation, and zero-shot task success across multiple benchmarks and real-world robot tasks. The model's architecture integrates world modeling with action generation through a shared transformer-based framework.
- arXiv Robotics — research abstractsSignalGreenWAM: A World Foundation Model for Robotics
This paper introduces GreenWAM, a world foundation model for robotics that serves both world modeling and policy generation roles. It achieves top-tier performance on WorldArena 1.0 and ranks among the top entries in RoboTwin 2.0. The model is adapted for real-world humanoid robots and distilled into a single-step policy for real-time control, with controlled experiments highlighting the benefits of robot specialization and maintaining world-model modes during policy adaptation.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).