FASTER

FASTER is a simple yet effective framework for fast action sampling and immediate reaction in real-time flow-based VLAs.

FASTER FASTER: Rethinking Real-Time Flow VLAs Yuxiang Lu 1,2 , Zhe Liu 1 , Xianzhe Fan 1 , Zhenya Yang 1 , Jinghua Hou 1 , Junyi Li 1 , Kaixin Ding 1 , Hengshuang Zhao 1 1 The University of Hong Kong 2 ACE Robotics arXiv Code Real-time reaction in VLAs is constrained not only by inference latency, but also by how action chunks are generated and executed. FASTER introduces a new paradigm for fast action sampling under asynchronous execution. By compressing the sampling process for immediate reaction into a single step, FASTER achieves 10x acceleration over π 0.5 and X-VLA, enabling real-time

saved by

• Anya Singh

related reading

• [2510.04371] Speculative Actions: A Lossless Framework for Faster Agentic Systemsarxiv.org
• 45d74e190008c7bff2845ffc8e3facd3-Paper-Conference.pdfproceedings.iclr.cc
• Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Modelsarxiv.org
• Precise Manipulation with Efficient Online RLpi.website
• The First Fully General Computer Action Model | blogsi.inc
• Real-Time Action Chunking with Large Modelspi.website
• State of Vision-Language-Action (VLA) Research at ICLR 2026 – Moritz Reussmbreuss.github.io
• Real-Time Action Chunking with Large Modelsphysicalintelligence.company
• Real-Time Execution of Action Chunking Flow Policiesarxiv.org
• FAST: Efficient Action Tokenization for Vision-Language-Action Modelsarxiv.org
• Interaction Models: A Scalable Approach to Human-AI Collaboration - Thinking Machines Labthinkingmachines.ai
• [2603.16666] Fast-WAM: Do World Action Models Need Test-time Future Imagination?arxiv.org