AWS re:Invent 2025 - Designing local Generative AI inference with AWS IoT Greengrass (DEV316)
AWS Events
12 views • 7 months ago
Video Summary
Short Highlights
- Keywords Physical AI, local inference, cloud inference, generative AI, robotics, latency, AWS IoT Greengrass, VLM, edge AI, model updates, Raspberry Pi, Docker, containers
Key Details
- Short Keypoints
- Physical AI involves machines sensing, deciding, and acting in the real world, moving AI beyond screens into physical interactions.
- Local inference is crucial for Physical AI due to responsiveness (less than 100ms latency), autonomy, and collaboration needs, especially in robotics.
- Cloud inference demonstrated a latency of 500-600 milliseconds for robot arm control, significantly impacting real-time performance.
- Keeping AI models updatable is a core idea in the physical AI era, as today's models can be outdated tomorrow.
- AWS IoT Greengrass is presented as a solution for delivering and managing AI model updates on edge devices, turning hardware challenges into an updatable software platform.
