Real-Time Action Chunking with Large Models

physicalintelligence.company · 1,732 words · saved by 1 readers

Unlike chatbots or image generators, robots must operate in real time. While a robot is “thinking”, the world around it evolves according to physical laws, so delays between inputs and outputs have a tangible impact on performance. For a language model, the difference between fast and slow generation is a satisfied or annoyed user; for a vision-language-action model (VLA), it could be the difference between a robot handing you a hot coffee or spilling it in your lap. While VLAs have achieved promising results in open-world generalization, they can be slow to run. Like their cousins in language and vision, these models have billions of parameters and require heavy-duty GPUs. On edge devices like mobile robots, that adds even more latency for network communication between a centralized inference server and the robot 1 . To build a real-time system with VLAs, we are going to need some form of asynchrony: that is, we must let a model think about its future actions while executing a previou

Real-Time Action Chunking with Large Models Real-Time Action Chunking with Large Models Published June 9, 2025 Email research@physicalintelligence.company Kevin Black, Manuel Y. Galliker, Sergey Levine Paper Follow-Up Paper Update (Dec 8th, 2025): we released a follow-up paper describing a training-time version of RTC, which we used in the π * 0.6 espresso-making demo ! Loading… Unlike chatbots or image generators, robots must operate in real time . While a robot is “thinking”, the world around it evolves according to physical laws, so delays between inputs and outputs have a tangible impact o

Real-Time Action Chunking with Large Models

related reading