Stanford's Real-Time EXPO-FT Makes VLA Robots Real-Time

Stanford researchers including Chelsea Finn released Real-Time EXPO-FT, a reinforcement learning fine-tuning framework that enables slow pretrained VLA policies to run in real time via an asynchronous split that continuously generates candidate action chunks, letting robots keep pace with the physical world.

2026-09-22 ~ 2026-09-24 · 2 related posts