Physical Intelligence releases FAST robot action tokenizer for efficient VLA training
Evidence notes
- Physical Intelligence published FAST, an efficient robot action tokenizer designed to connect high-frequency robot action chunks to autoregressive transformer training.
- FAST was trained on one million real robot action sequences and supports faster pi0-FAST policy training while retaining dexterous-control capability.
- Tokenizer and dataset infrastructure are part of the research substrate behind scalable robotic foundation models.
Company context
Physical Intelligence develops foundation models and learning algorithms for robots and other physically actuated systems. Its π-series models are designed to generalize across robot embodiments, tasks, and environments using robot data, language instructions, vision-language-action training, reinforcement learning, and multimodal context conditioning.