A 14k-star Chinese encyclopedia for robots that actually learn
A beginner-built knowledge base mapping the entire embodied AI stack—from simulation to VLA models—because the field needed a sane entry point.

What it does
Embodied-AI-Guide is a Chinese-language knowledge base and curated index for embodied intelligence research. It organizes papers, tools, tutorials, and community resources across algorithms (VLA, RL, IL, LLM+planners), simulators, benchmarks, datasets, and hardware. The project explicitly targets newcomers, offering a one-week hands-on track through the RoboTwin 2.0 simulation platform to train and evaluate an ACT policy on a manipulation task.
The interesting bit
The guide is built by self-described beginners for beginners—a rare admission in a field drowning in expert presumption. Rather than dumping paper links, it structures the “boring” infrastructure layer (simulators, benchmarks, data pipelines) as first-class citizens, arguing these determine reproducibility ceilings more than any single model architecture.
Key highlights
- Curated paper lists spanning VLA, affordance learning, humanoid robotics, and efficient VLA variants, with explicit links to top-tier venues (CoRL, RSS, ICRA, etc.)
- Hands-on tutorial requiring ~16GB GPU, walking through data generation → ACT training → evaluation in RoboTwin 2.0/SAPIEN
- Community ecosystem index: Chinese-language blogs, WeChat public accounts, Xiaohongshu influencers, and lab directories
- Annual trend summaries (e.g., 2025 embodied VLA retrospectives) to track how “demo” transitions toward generality
- Active Lumina community with contributor graph visible, 14k+ stars as of late 2025
Caveats
- Primary language is Chinese; English speakers get partial utility from paper links and code references
- The hands-on track depends on external RoboTwin 2.0 platform stability, not this repo’s own code
- README truncates mid-section (“Control -”), suggesting ongoing reorganization despite the January 2026 “reorganization complete” notice
Verdict
Worth bookmarking if you’re entering embodied AI from a Chinese-speaking context or need a structured map of where VLA, simulators, and data pipelines intersect. Skip if you want runnable frameworks; this is a curated index, not a codebase.