Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Research & technology· 日本語で読む

Physical Intelligence Releases RTC, a Real-Time Execution Method for VLAs; Maintains Performance With +200ms of Inference Latency

Physical Intelligence announced on June 9, 2025, Real-Time Chunking (RTC), an algorithm that enables real-time execution of diffusion- and flow-based VLAs without changes to training. RTC addresses discontinuities between chunks as an inpainting problem and reduces execution time compared with synchronous inference. The company says performance does not decline even with an artificial delay of +200ms.

What happened

Physical Intelligence announced on June 9, 2025, Real-Time Chunking (RTC), an algorithm that enables real-time execution of diffusion- and flow-based VLAs without changes to training. RTC addresses discontinuities between chunks as an inpainting problem and reduces execution time compared with synchronous inference. The company says performance does not decline even with an artificial delay of +200ms.

Why it matters

For robotics developers seeking to run diffusion- and flow-based VLAs in real time without retraining, RTC offers a way to eliminate pauses between inference runs. Physical Intelligence says the method reduces execution time compared with synchronous inference and showed no change in performance with an artificial delay of +200ms, providing concrete conditions for comparison in real-world deployments where latency can be an issue.

Sources

Translated from the Japanese original: /news/physical-intelligence-robotto-no-riarutaimu-jikkou-wo-kanoun-3052e0. The Japanese version is authoritative if they differ.