Space clouds and onboard AI: Future of satellites ready to lift off

While space researchers around the world are scrambling to track the growing number of satellites above us, Dr Carl Seubert wants satellites to track themselves.

The Chief Research Officer at SmartSat Cooperative Research Centre (CRC) says satellites of the future will make their own decisions, would have AI onboard, and even work within their own ‘cloud’ system.

While these claims may sound a bit ‘out of this world’, a research project called SCARLET (SpaceCraft Autonomy Research Laboratory) launched in June with a goal to try to make it happen.

“Australia has a great software, technology and AI background. It’s now about taking that and lifting it up into space,” Seubert, who worked at NASA for almost a decade, told Cosmos.

“When I came back to Australia I had this drive and passion to keep controlling spacecraft autonomously.”

Once they make it into orbit, satellites and other spacecraft are relatively simple technology.

They might take images or collect data, but that information needs to be beamed back down to Earth and retrieving the data isn’t quick. Because of the way data is transferred back to Earth, and because satellites are constantly moving, it can take up to a day to get the information required.

Similar to slowly sucking a drink out of a cup, Seubert calls this trickle of information ‘through a straw’, and if you need the information straight away the delay can have dire consequences.

“A good example of that is early fire or smoke detection,” he said.

While data and images can be removed off satellites to check if there is smoke or fire manually, analysing those images on the satellites could allow faster detection.

One of SmartSat’s recent projects called the Kanyini satellite looked into this using an AI onboard algorithm, with an imager to be able to take and then analyse it all onboard.

Kanyini “aims to demonstrate early smoke detection using AI processing on hyperspectral images where you use the spectral bands to differentiate early smoke compared to other signals that might look similar – like clouds, fog or mist,” says Seubert.

“You want to do that quickly and you want to get that information off the spacecraft quickly. So, you can then ultimately make a decision on the ground and send resources to act on it.”

The delay also impacts how satellites are directed in space.

In most cases, satellites in low-Earth orbit need to be individually tracked and given manual movement information to get out of the way of passing satellites or other junk that might be in their path.

“When we fly spacecraft, we truly operate them as individual elements. We know exactly where a spacecraft is, we control it, command it, we know exactly when we’re going to get the telemetry,” says Seubert.

This is fine if we have just a few satellites, but with internet constellations like Elon Musk’s Starlink proposing 42,000 satellites, it starts to get crowded.

“How do we now control that constellation? You need to use autonomy to do that,” he adds.  

“Autonomy to support space traffic management is very much one of our agendas.”

The final – and most complicated – type of future satellite is the ‘space cloud’ – a series of interconnected satellites with barely any human involvement.

In this space cloud, different satellites can undertake different tasks – one might image, while another might process and a third (or more) feeds the data back to the ground. All of which are talking to each other and avoiding collisions. 

“We perhaps have a constellation of small spacecraft, they’re all connected through radio frequency or through optical links and they’re sharing information, sharing data,” says Seubert.

“The end user might not have anything to do with the spacecraft or know what’s even up there. Instead they’d just be tasking and requesting information or a service.”

Of course, the journey from today’s satellites to the space cloud might not be smooth sailing. There are still limits on what processors can do, and all processes need to be specially designed for space.

Working out how to connect the satellites and the technology is another issue for the future.

But Seubert is optimistic that these problems can be overcome.

“To be honest, it’s quite futuristic, but we need to start pushing our technology towards this. My vision is very much about well, let’s assume we’ve got connectivity. Let’s assume we’ve got better processing. Let’s assume we’ve got the space cloud,” he told Cosmos.

“Now let’s build the smarts to start developing the actual information extraction that we want out of it.”

Please login to favourite this article.