靈波機器人模型先預判再動手. 本文保留已核驗事實，並以產品、基礎設施與治理脈絡重新整理。
發生了什麼
靈波機器人模型先預判再動手 is the current news peg. The verified record centers on these points:
- LingBot-VA 2.0 combines video prediction and action generation in one sequence
- the published target is single-card 150Hz real-time inference
- GitHub tables list RoboTwin 2.0 and LIBERO benchmark results
- real-world tasks include breakfast, screw retrieval, tube insertion, unpacking, shirt folding and pants folding
為何重要
更大的產業訊號是 embodied AI is moving from perception modules toward models that predict how action changes the environment.
接下來看什麼
接下來要看的重點是 WAIC demos, reproducible community tests and stability on long, messy manipulation tasks.
參考來源：QuantumBit, Beijing News Shell Finance, Robbyant GitHub README, Hugging Face model card, CocoLoop.