FAST: Efficient Action Tokenization for Vision-Language-Action Models

By Karl Pertsch, Kyle Stachowicz, Brian Ichter, Danny Driess, Suraj Nair, Quan Vuong, Oier Mees, Chelsea Finn, Sergey Levine. Robotics, 2025

551 citations recorded by the source corpus.

Autoregressive sequence models, such as Transformer-based vision-language action (VLA) policies, can be tremendously effective for capturing complex and generalizable robotic behaviors. However, such models require us to choose a tokenization of our continuous action signals, which determines how the discrete symbols predicted by the model map to continuous robot actions. We find that current approaches for robot action tokenization, based on simple per-dimension, per-timestep binning schemes, typically perform poorly when learning dexterous skills from high-frequency robot data. To address this challenge, we propose a new compression-based tokenization scheme for robot actions, based on the discrete cosine transform. Our tokenization approach, Frequency-space Action Sequence Tokenization (FAST), enables us to train autoregressive VLAs for highly dexterous and high-frequency tasks where standard discretization methods fail completely. Based on FAST, we release FAST+, a universal robot action tokenizer, trained on 1M real robot action trajectories. It can be used as a black-box tokenizer for a wide range of robot action sequences, with diverse action spaces and control frequencies. Finally, we show that, when combined with the pi0 VLA, our method can scale to training on 10k hours of robot data and match the performance of diffusion VLAs, while reducing training time by up to 5x.

View Paper: https://www.semanticscholar.org/paper/3880f5bad862ba1b18f4f8ec060038b326b118ed

Citations in the corpus

• Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success (2025, unknown): “Recent works by Belkhale and Sadigh [2] and Pertsch et al. [36] improve VLA efficiency through new action to-kenization schemes, using vector quantization or discrete co-sine transform-based compression to represent action chunks (sequences of actions) with fewer tokens than simple per-dimension…”
• π0.5: a Vision-Language-Action Model with Open-World Generalization (2025, unknown): “However, as shown in [64], VLA training can be much faster when actions are represented by discrete tokens, particularly when using a tokenization scheme that is efficient for compressing the action chunks (e.g., FAST).”
• SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Model (2025, unknown): “Recently, several studies [34, 7, 30, 31, 71, 33, 70, 51] propose to build generalist robot policies by extending pre-trained VLMs with ability to robot action generation.”
• SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics (2025, unknown): “We use a dataset (Kim et al., 2024; Pertsch et al., 2025) 1 containing 1,693 episodes covering all tasks, and evaluate with 10 trials per task, reporting average success rates based on binary completion criteria.”