TeleAI Transmits Video as Tokens to Satellites
量子位 · wechat · 2026-07-18
TeleAI Proposes "AIFlow"
China Telecom's AI Research Institute (TeleAI) launched AIFlow at WAIC 2026. The system understands and encodes video and voice into Tokens on the device side, transmits them via satellite, 5G, and other links, and uses generative models to reconstruct the content on the receiving end. The core concept is to trade computing for bandwidth, enabling smart terminals to transmit information in narrowband, disconnected, or long-distance scenarios.
Deployed in Multiple Extreme Scenarios
The article notes that this solution is already used on the front lines of flood prevention, with over a thousand devices deployed across 31 provinces in China. It also enables small drones to connect directly to satellites for HD video transmission, cross-city teleoperation of robots, and remote wireless HD video transmission in natural waters.
Key Data and Goals
- Audio: Enables voice calls and music transmission at 0.266 Kbps, claiming a roughly 500x reduction compared to traditional codecs
- Video: Can transmit and reconstruct 1080P@30fps video at lower bitrates
- Storage: For 100 million camera video resources, storage space can be reduced to 1/40th of traditional solutions by retaining tokens on demand
The article summarizes this approach as moving "from bitstreams to token streams," aiming to allow AI capabilities to flow seamlessly between drones, robots, cameras, vehicles, and underwater devices.
More from Embodied
- LIBERO-MAX: Robot Policies Drop 11–26 Points When the World Changes Mid-Task — drmapavone · 2026-10-27
- Photos from ROSCon UK: BBC's robotic whale, da Vinci surgical robot, PR2 and Baxter — kamathsblog · 2026-10-27
- Agility's Digit robot to expand into two more GXO US warehouses from early 2027 — jonstephens85 · 2026-10-27
- Real2Sem2Real turns one phone video into 2,000+ robot demos, matching 150 teleop demos — Scobleizer · 2026-10-27
- Wayve CEO argues end-to-end deep learning should power driving, safety and world models — alexgkendall · 2026-10-27
- Mundo turns a few room photos into an explorable, simulatable 3D world — at 1.5 FPS — mark_k · 2026-10-27