Drones were where the money was, and some of the TinyOS research carried straight over, so it was a pivot of the idea rather than a clean break. D-Watch became an autonomous drone surveillance system: you plot an area on a map and a swarm goes and surveys it. My grad research was in computer vision, so the two halves fit together.
The technical core is the part I would still defend. The best vision model available then was Yann LeCun's convolutional neural net, and I deconstructed it so that it ran across three tiers. The drone, the hub and the cloud, as though it were one system, at low latency, with staged outputs at each tier at increasing confidence. That was 2015.
The other decision I would defend is privacy. One of the biggest regulatory pressures on camera drones across the US and the EU was privacy, and rather than wait for the rules to arrive we designed the prediction system to be privacy-conscious from the start. That is the decision that later got me hired by Google.
It did not start with drones at all. I left Microsoft to build TinyOS, a federated computing company: a decentralised operating system running across devices large and small, sharing resources between them, so a device short on memory could draw on one that had it. Nobody would invest in that. It is a reasonably ordinary idea now.
The ending. Regulation arrived and restricted how high drones could fly and what they could be used for, and it killed almost every drone company in the US. It killed ours. It felt horrible. It felt like I was just doing everything wrong, and nobody really cared about the innovation we believed we had produced. Building ahead of the moment means sometimes the moment never comes.
Closed by regulation, along with most of the US drone industry. The instincts stuck: inference at the edge, and treating where computation happens as a privacy decision rather than only an engineering one.