Page number 18
AI in rbit Fintan Buckley , Trinity graduate, co-founder and CEO of Irish space-tech company Ubotica, explains how they are revolutionising satellite technology using AI F inding algal blooms before they can create ‘dead zones’ and harm aquaculture. Locating the ‘dead vessels’ involved in human trafficking, drug smuggling, illegal fishing. On land, assessing irrigation problems, tracking erosion and measuring the impact of a quake or a flood: all of these are problems that Irish space-tech startup Ubotica is helping solve by allowing machines to ‘see’ the earth from orbit more efficiently. They do that using Artificial Intelligence (AI) which is – ground-breakingly – built into hardware that is incorporated directly into the satellite and launched into orbit. The AI image processing happens right away as sensors capture the images of our planet. This is a tricky maneuver. Commercially valuable AI analysis involves increasingly massive processing and storage capacity, neither of them practical to put on a payload to launch into space. So why is this also a wise move? The AI acceleration on earth is associated with rising costs in energy and computing power, as well as questions of data sourcing, both to train the AI models and to feed to them once they’re mature, for profitable output: AI thrives on fresh data, and access to data in real time, together with the ability to extract meaning from it, can make one’s model an indispensable asset, beating competitors. Satellite imaging data of earth is some of the Old: Centralised Intelligence – simple satellites New: Edge-AI Intelligence – smart satellites
Page number 19
freshest, yet also the bulkiest data around. There are cost and latency hang-ups involved with transmitting it back to earth, with storing it, and with the time elapse between the capture and the processing. Ubotica CEO Fintan Buckley says: ‘A lot of the satellites that are flying now are "imagers" that are recording the Earth, and they are creating data mountains, contributing to issues around climate change, with the data centers which are required to store it all. And so what we are doing is pushing the processing of this image data from the ground onto the satellite.’ Ubotica’s CogniSAT platform therefore enables autonomous decision-making directly on satellites: making the machine ‘see’ – machine or computer vision – means combining the camera or sensor with AI to allow it to inspect, analyse and act independently. ‘There's a number of ways in which this wins,’ Fintan adds. ‘First of all, if you're in a country like Ireland, where we encounter so much cloud cover, if a satellite takes an image on a day when it's cloudy, there's no point in even downloading that image to Earth. So you use AI to analyse the image and say, "discard it, don't bother sending it to Earth".’ With the dark vessels application (already deployed in Jeddah, Galveston and Los Angeles) a problem with transmission of raw data would be that the suspect vessel, by the time authorities have identified it, will have moved on. Bringing identification closer to real time should be irresistible for clients wanting to pinpoint the smugglers. What Ubotica is able to offer now is possible thanks to lower costs in space tech and in hardware manufacturing, with collaborators coming from the US and China. ‘We are in an age where Elon Musk’s SpaceX has commoditised and simplified the ability to put satellites into space, opening the path to using what we call Commercial Off-The-Shelf (COTS) technology. We take consumer-grade technology for doing AI processing and deploy that on spacecraft.’ Hunting that hardware has led Ubotica to the East, where a manufacturing partner has worked with them over the past five years. Fintan began to visit the country, and was impressed with Shanghai and Hangzhou, the tech-hub and hometown of Alibaba and the vast ornamental West Lake. Left to right: Co-founders Aubrey Dunne, Fintan Buckley, and John Bourke.
