World Models
GE-Sim 2.0: a closed-loop video world simulator for manipulation
A talk on AgiBot Genie's GE-Sim 2.0 — turning a video generator into an environment robots can actually train and test in, via a ...
Blog
Notes & talks.
Durable notes and recorded walkthroughs — organized by topic. Each card is marked Note for written pieces or Talk for video explainers.
Embodied AI
Embodied AI, full stack: body, brain, and safety
A three-part survey of embodied intelligence — robot bodies, simulators, perception and grasping; world models; and the duality o...
Autonomous Driving
Full-stack autonomous driving: from sensors to control
A talk through a complete AD stack — dual-brain domain control, multi-view + LiDAR perception, BEV and occupancy, online mapping,...
Generative Models
One-step generative models: the MeanFlow line, paper by paper
A four-paper walkthrough of He Kaiming's group compressing diffusion's hundreds of steps down to one — MeanFlow, improved iMF, dr...
World Models
OmniDreams: a real-time generative world model for driving
A talk walkthrough of NVIDIA's OmniDreams — a causal autoregressive DiT world model that closes the loop between policy and simul...
Scene Reconstruction
从 3DGRUT 到 ArtiFixer3D：一个“先重建、再生成、再蒸馏回 3D”的闭环
Reading ArtiFixer (SIGGRAPH 2026, NVIDIA Toronto AI Lab) — an interactive 3D logic atlas of the reconstruct → generate → distill-...
Embodied AI
从互联网到物理世界：具身智能的数据革命从哪里开始？
一份关于具身智能数据采集的工作笔记——在保真·接触、规模·成本、跨本体三者难以兼得的约束下，遥操作、仿真、UMI、Ego 各自下了什么赌注。
Embodied AI
机器人怎么"练"出来：在真实世界里，还是在想象里
以 RECAP / π*0.6（真实世界做 RL）和 RISE（世界模型内做 RL）为两条主线，讲清楚 VLA 为什么单靠模仿学不好、两种破局路线各自的本质与代价。