Colm Gorey | Silicon Republic | 18 Jun 2019
2 min read
‘Visionary’ start-up aims to bring low-power AI to the edge of space
Irish start-up Ubotica is finding ways to bring advanced AI vision systems to Earth and space alike.
Artificial intelligence (AI) comes in many forms. Whether it’s Google-owned DeepMind trying to develop a board game master, or Amazon trying to perfect its Alexa personal assistant, all are racing towards a very different future.
However, one facet of the technology that you might miss at first glance is computer vision. Away from the vast systems you’ve seen placed on top of prototype autonomous cars filled with tech, you have something called edge-based systems.
These are attempting to take AI systems to the smallest of devices that can see what’s in front of them, gather data and react all on their own. So, for example, an internet of things (IoT) sensor placed in a factory that tracks vehicles driving around a warehouse.
This is one area in which Dublin-based start-up Ubotica is aiming to flourish, but it is also pushing this technology to the edge of space and beyond.
Myriad of uses
In November of last year, the company worked with the European Space Agency (ESA) to blast a potentially space-bound AI chip with powerful radiation. This chip, the Myriad 2, was created with Irish expertise as its architecture was originally developed by the AI start-up Movidius, which was snapped up by Intel in 2016.
Ubotica, founded in 2017, has strong ties to Movidius, with many of its senior members once being an integral part of its management. With prior experience using the Myriad device, the company’s technology director, Aubrey Dunne, said it seemed a natural fit to keep working with what is now Intel’s low-power, artificially intelligent pride and joy.
“When Movidius was acquired by Intel, we saw an opportunity to focus on bringing the ability of this chip to different applications and essentially to products or application spaces that haven’t been addressed necessarily with edge devices,” he said in conversation with Siliconrepublic.com
When asked whether there were other alternative chips for Ubotica to develop its edge-based AI vision technology, he said what made the Myriad the best option was that it was the first chip designed from the ground up for vision processing.
“You see that other developers of similar types of chips – or even other larger chip developers – are trying to move into this space. But they take their existing technology and try to adopt their existing chip architecture to the low-power case, and ultimately that will always be a little more power-hungry than when you designed from the ground up.”
AI in high places
Additionally, within the space sector that Ubotica is already working in, the Myriad is one of the best commercial chips of its kind at withstanding the rigours of space and the freezing temperatures that come with it. An example Dunne gave of where a combination of the start-up’s technology and Intel hardware could flourish is in Earth observation.
Satellites that orbit the Earth to capture images of the surface below typically store huge quantities of images before sending them all back in one data dump at a pre-determined point. This could include countless images of cloudy skies that serve no use to scientists on the ground.
What Ubotica is trying to build, however, is a smart system that can determine whether an image is just cloud cover or something worth sending back.
“Now, you’re reducing the amount of data that you’re sending down to Earth, which means it’s using less power, it costs less and you’re also making much more efficient use of your downlink bandwidth,” Dunne explained.
Back down to Earth
The Myriad 2 chip, he added, is crucial for working with the lowest amount of power necessary, unlike power-hungry, bulky sensors seen in technologies such as autonomous cars. Not only that, but it must withstand being bombarded by cosmic radiation, as the ESA tests have shown.
Yet space isn’t the only application for AI vision, with promising developments in the ability to diagnose diabetic retinopathy – one of the world’s leading causes of blindness – also among Ubotica’s repertoire. Using its AI vision technology, doctors could soon scan a patient’s eye and support a diagnosis of the disease as a second opinion.
“That image is analysed by our neural networks and, with a degree of certainty, it can decide whether a patient should be referred for further check-up, and that information comes directly out of the camera.”
This makes it more efficient at processing data, but also much more secure from a data privacy perspective and compliant with GDPR, in the EU at least.
Looking to the future, the company is also aiming to help other start-ups use the Myriad chip, as seen with the recent inaugural Intel Movidius Edge AI Incubator Open Day at Talent Garden Dublin, featuring 10 start-ups from across Ireland and Europe.
“We were supporting that programme from a technical viewpoint, so it was actually very interesting because it allowed us to see some of the interesting ideas that people are coming up with around an edge-based vision,” Dunne said.
Published by Silicon Republic – read it here