Lightning Talk: An Efficient Automotive Safety Support System Utilizing Hardware...
Lightningat Open Source Summit Japan + AI_dev: Open Source GenAI & ML Summit + Automotive Linux Summit 2025 · 2025-12-08..12-10·
summary
Proposes moving camera and sensor streams from roadside units into centralized GPUs over an Open APN with RDMA to remote GPUs, claiming end-to-end optical paths that give about 100× power efficiency and 125× transmission capacity versus conventional links and latency cut by orders of magnitude. Targets concrete use cases: road condition surveillance and blind‑spot prediction for lane merging. Architecture replaces thousands of edge GPUs with a hardware‑accelerated central pipeline, promising lower cost and energy; proof‑of‑concept experiments are planned.
AI-written summary · about
related talks
21mApidays Virtual Helsinki - Merja Kajava, Aavista
19mAI Trends in Intelligent Transport Systems
57mBuilding GPU-accelerated multi-agent apps
22mExtending European Strength with Embedded AI
57mWriting Safety Critical Automotive Software for High Perf AI Hardware
12mKeynote: Accelerating Autonomous Driving Network via Industry Collaboration
50mThe democratization of tomorrow’s technology
12mEfficient Workflow Serving for Diffusion Models with Many Adapters