Meet Darcy, a Coral- powered platform that streamlines cloud-to-edge AI

A sanitation truck makes its way through an otherwise quiet neighborhood of a mid-size American city just after dawn. The truck stops, and a worker jumps off the back and heads to the curb for a garbage container, but the truck is far from idle. So far, cameras mounted on the side have spotted — and software has logged — five potholes on this block alone that need attention from city engineers.

The vehicle is doing double-duty for both the sanitation and highway departments, thanks to onboard AI, enabled by Coral.

This is just one use case envisioned by the founders and engineers of Edgeworx, a startup company based in Silicon Valley, California and New Zealand. Its Darcy product, a device the size of a paperback novel, provides everything developers and even a one-person IT department need to deploy privacy-preserving, EdgeAI applications.

"We're an edge-cloud company," says President and co-founder Farah Papaioannou. "We want to enable ML developers to run any AI applications they want to at the edge." The company wants to make deploying and managing AI applications as easy at the edge as it is in the cloud, today.

Darcy and its user-friendly software make the potentially challenging task of standing up AI applications much easier for the likes of city administrators, school districts, and even farmers looking for ways to monitor remote fields and orchards for areas that need attention. The idea is that Darcy’s all-in-one platform that includes video, thermal and sound sensors and user-friendly software can bring machine learning within reach of non-specialists.

Under the hood

Each Darcy packs a powerful AI punch with the help of multiple sensors, an embedded ARM CPU, built-in Wifi, BLE, hardware-based security, and the Coral Edge TPU™. Darcy currently uses two Coral USB Accelerators, and Edgeworx plans to build as many as four into a single device.

Multiple AI accelerators in one device might seem like overkill, but Papaioannou explains that it makes the device more responsive thanks to the ability to detect several events or objects simultaneously.

"We run Darcy with multiple Coral co-processors so we can run multiple machine learning models at once," Papaioannou explains. "You don't have to just solve for one thing. On campus, maybe I want to see if there's a fire. I want to detect school shootings. I want to see if a fight breaks out. I want to be able to check for all of these things at once so I can react to them faster."

That powerful capability to run a separate model for each situation to detect is a prime reason the Edgeworx team selected Coral for Darcy. And although one Edge TPU is fast enough to run more than one model in real-time, additional Edge TPUs allow you to pipeline or parallelize much larger neural networks. Edgeworx also provides a full cloud backend for the fleet management of Darcy devices and the deployment and lifecycle management of AI models.

Protecting schools and workplaces

Valley Christian Schools is a K-12 private school with three campuses in San Jose. The schools had to meet stringent California regulations around reopening following the worst of the COVID-19 pandemic. Darcy provided the answer with automated screening for students and staff as they entered the building each morning.

Now, students enter buildings safely with the help of friendly-looking kiosks powered by Darcy, helping avoid the spread of covid and keep schools open. Todd Baldwin, Director of Operations at the school affirms that Darcy has made life easier for him and those he supports. “The Darcy system has been super easy,” he says. “It’s been very useful in getting our school reopened.” When students walk into a school building, they present a QR code on a smartphone app to a kiosk with Darcy built in. Darcy scans that, uses a thermal camera to screen for fever and a video camera to check for a face mask, all in under 100ms. With no information processed in the cloud, all student data stays on the device, preserving privacy.

Darcy made it easier for the school to deploy an AI solution because setting it up requires no specialized skills. “There are a lot of complexities in deploying a production AI solution in the field that a developer would have to think about without Darcy,” Papaioannou says. “We make those things go away so they can just focus on their business use case.”

Edgeworx is also helping companies develop the means to check employees onto company buses and watch industrial plants for hazardous conditions, such as fires and equipment malfunctions that might endanger employees.

And there's the local government considering AI-powered cameras for its vehicles. "Buses and garbage trucks are already driving around, covering every inch of the city," Papaioannou says. "They want to have AI on them so they can identify infrastructure issues like potholes and signs that need replacing." Vehicles would upload data about infrastructure that needs a closer look by municipal engineers when they got back to their garages.

Next on the company's roadmap are dedicated Darcy models for different environments, including bandwidth-constrained farms and in the extreme environment of factories and city streets.

"Ultimately, what we deliver is the fastest time to deploy business AI at the edge," Papaioannou says. "I want to make it easy for developers and system integrators or value added resellers who have these specific use cases in mind to go in and rapidly build solutions for themselves." In other words, Edgeworx is democratizing Edge AI with a customizable plug-and-play system for designing and deploying complex AI solutions — solutions previously out of reach by all but highly specialized experts starting from scratch.

If you're interested in how Coral can help you build your next product, please contact us.