Visual Input
Robot Cameras (Images / Video)
Generative Learning of World (GLOW)
KnowinDream
Generative 3D World Data Engine
Generates higher-quality embodied data distributions equipped with rich 3D structure.
Language
User Text Instruction
Robot State & Execution Feedback
Proprioception / Rollout Traces
EMBODIED FOUNDATION MODEL
KnowinBrain
（Post-trained for Action CoT）
Autoregressive CoT aligned with action token generation.
ACTION TOKENIZER
KnowinAct
(Pre-trained Action Decoder)
The action token space is growing dynamically.
ACTIONCHUNK
KnowinWorld
Interactive World Dynamics Sandbox
Action rollout
Robot Execution
Action chunks are converted into smooth arm motion locally.
Embodied Alignment
World Dynamics World-grounded rollout
Active Exploration Skill composition
Interactive Reasoning Brain-Act consistency
RLHF