SEG-JPEG: Simple Visual Semantic Communications for Remote Operation of Automated Vehicles over Unreliable Wireless Networks
New AI encoding technique overlays color-coded road users on grayscale video to slash data needs.
Deep Dive
Researchers from the University of Warwick and others developed SEG-JPEG, a visual semantic communication system for remote vehicle operation. It encodes detected road user segmentations as color highlights on low-res grayscale imagery, reducing required data rates by 50% compared to conventional methods. This enables a median glass-to-glass latency below 200ms on networks under 500kbit/s, allowing reliable remote control of automated delivery vehicles over constrained public 4G/5G networks.
Why It Matters
Enables scalable remote operation of autonomous vehicles without requiring expensive, dedicated high-bandwidth network infrastructure.