Skip to content
Rex St. John
Dec 2019Armsummit

Arm AIoT Dev Summit

Two days at the Computer History Museum covering IoT from edge to cloud, intelligent and autonomous machines, and AI from development through deployment. Half on stage, half hands-on with hardware in the room.

Numbers

500+
Developers

Problem

AIoT was being discussed in two separate conversations that never met: cloud people talking about models, and embedded people talking about boards. A developer trying to deploy intelligence at the edge needed both and got whichever conference they picked.

Solution

Two days at the Computer History Museum in Mountain View, 2–3 December 2019, covering IoT from edge to cloud, intelligent and autonomous machines, and AI from development through deployment.

Half on stage, half hands-on with hardware in the room. Every attendee went home holding a board.

Analysis

I split the programme deliberately. The keynotes and panels set the frame: the latency arithmetic behind the case for edge compute, the cost table showing which robotics platform a hobbyist can start on, Ciira wa Maina on deploying AI where bandwidth and power are the binding constraints. Then everyone went into a room and built.

Free hardware for every attendee was the expensive line on the budget. I defended it because a developer who has to buy a board before evaluating it mostly does neither.

Result

More than 500 developers over two days.

My favourite moment was unscheduled: people in the lounge between sessions, assembling a drone frame on a coffee table.

Photos

  • A packed workshop room with every table covered in open laptops and development boards
    The half of the programme that mattered most. Laptops open, boards out, everyone building the same thing at once.
  • A three-person panel on stage beneath a slide reading "Deploying AI Solutions in Africa" by Ciira wa Maina of Dedan Kimathi University and Data Science Africa
    Ciira wa Maina on deploying AI in Africa — the session that most changed how people in the room framed constraint.
  • A speaker beside a slide comparing types of DIY robocars across computer vision and deep learning approaches, with hardware and cost for each
    A full cost breakdown of every DIY robocar platform, from a $90 OpenMV racer up. Price is the adoption curve.
  • A speaker on a dark stage presenting a latency breakdown slide headed "This is a Cloud-scale Problem", totalling 775 milliseconds per event
    775 milliseconds per event, round trip. The whole argument for edge compute in one slide.
  • A second workshop room mid-session, attendees at laptops with presenters working the room
  • Four attendees gathered around a laptop working through a problem together during a break
  • Attendees in the lounge assembling a drone frame on a coffee table scattered with components and Arm stickers
    Hardware in the lounge. Nobody scheduled this one.

Sources

AIoTSemiconductorsAIRobotics