Allyson Park
With an increasing number of commercial satellites, military constellations and all their data flowing through space, government and industry operators said they are looking to artificial intelligence to help them to handle ever-more complex missions.
AI software is enabling users to change a spacecraft’s capabilities while in orbit, making the satellites less reliant on hardware.
“AI doesn’t exist without software,” Aslan Tricha, vice president of automation and orchestration at ALL.SPACE, a maker of satellite terminals for the military, said during a panel at the recent Satellite 2024 Conference and Exhibition in Washington, D.C. But the satellite industry has traditionally centered on hardware. “We build everything. It’s fixed. It’s kind of deployed as is, and it has to last 20 years,” he said.
Tricha said going forward, “we need to think about software and how those changes need to happen on the satellite.”
Since satellites are static and hardware-reliant, once they are launched into orbit, changes can generally not be made. That could change with AI-enabled spacecraft, he said.
“Learning patterns means learning a pattern of behavior. So, for example, if the requirement is changing because we have a higher demand — peak demand or different requirement — then AI is able to change the hardware architecture to fit the requirement,” he said.
The European Space Agency and Intel were the first to publicly acknowledge sending an AI-enabled satellite to space in September 2020. The PhiSat-1 — with an Earth-observation payload designed to monitor polar ice and soil moisture — had Intel’s Movidius Myriad 2 vision processing unit chip aboard, which was not originally designed for the rigors of space, according to an Intel press release.
Computer chips must be hardened for the extreme radiation found in space, and the ones used on orbit can be technologically as much as 20 years behind chips used on Earth, the statement said.
Adapting the state-of-the-art Movidius Myriad 2 chip for the rigors of space was what made the mission a breakthrough, a European Space Agency statement said.
The chip was able to use AI to sort out the unclear imagery from the useful ones on board the spacecraft and send only the good data back to Earth.
“By only sending useful pixels, the satellite will now improve bandwidth utilization and significantly reduce aggregated downlink costs — not to mention saving scientists’ time on the ground,” the agency statement said.
Carlos Pedalino, vice president of product and head of Latin America business at ReOrbit, a manufacturer of small communications and Earth-observation satellites, said using AI to make satellites software-centric will also eliminate reliance on ground stations and reduce mission costs overall.
“What we are trying to do is reduce the human dependence on ground stations. The cost of the mission is not just the cost of the satellite,” he said. “You need to operate [it], and you need a 24/7 team controlling the satellite, so we are trying to reduce that in order to reduce the cost of the complete mission.”
AI is also helping commercial satellites interpret signals, sort through massive amounts of data, speed up analysis and weed out useless or unhelpful information, he said.
When you don’t have full visibility of a satellite that is capable of various maneuvers, the satellite itself needs to learn what is going on and “make the right decision” based on that existing understanding, Pedalino said.
For example, for remote sensing, optical cameras need to select the right target, then send all the relevant data the most efficient way possible, a process that is only made better with AI.
“This data flow management between satellites is managed by AI. All the traffic data between satellites is driven by this software,” Pedalino said. “We have a layer in the satellite, and then we have a second layer in the ground to continue working on the image to get all the input, the additional data on the images and on the image processing software as well.”
Satellite operators can use AI to their advantage to help manage operators’ cognitive overload. With the vast amount of data gathered by both commercial and military satellites, AI is helping sort through and analyze the information in real time through on-board processing.
Satellites “need to be able to disseminate what is true and what is false. What is the right information to act on versus what is not. AI can help with this,” Tricha said.
Two years after the ESA/Intel PhiSat-1 mission, Palantir and Satellogic launched an AI-enabled satellite. It incorporated Palantir’s Edge AI technology onboard Satellogic’s NewSat to process imagery data on orbit, “separating signal from noise in high-scale data to make the best use of limited bandwidth,” a Palantir statement said.
“Edge AI starts processing data upon capture, delivering actionable insights faster than traditional ground processing. For example, we can inform future data collection onboard … or selectively downlink pre-processed images based on their analytical value,” the statement added.
Tricha said: “We need to be able to look at how much of this processing we do at the edge versus how much of this processing we do in the command center or other places. If a drone is sending terabytes of image data that has to go over a very tiny satellite pipe into some data center … to be processed to then go all the way back, I think the advantage of time is lost.”
Industry is also looking to use AI to develop smart terminals on the ground that receive satellite transmissions and data, as well as come up with a “smart routing system” to manage and visualize commercial and military satellite movements, Tricha said.
He defined smart ground terminals as receivers that are “semi-aware of their environment,” house the appropriate data and can transmit and receive the information to and from multiple satellites from both commercial companies and the military.
These smart terminals must understand the signals coming from the data so “knowns” and “unknowns” can be differentiated, as well as communicate with multiple satellites and execute multiple missions at the same time, receive that data, then act on it, Tricha said.
If these terminals can communicate with low-Earth orbit, medium-Earth orbit and geostationary satellites at the same time as they link with allied satellites, then you “can kind of inform the other terminals in the constellation, and you can do all of this in a very, very confined and mobile environment,” he said.
With the number of satellites increasing, an AI-enhanced smart routing system is needed to keep track of orbits and movement, said Karen Florschütz, executive vice president of connected intelligence at Airbus Defence and Space.
“It’s getting a little bit more crowded, and we need a smart routing system. Who is on which path up there, and which one is a communication one, which one is an optical one, which one is a radar one?” she said.
“To actually be able to process all the information on the ground, it needs to all come together. And maybe we don’t need, in the future, like 10 different ground stations, but maybe we need one or two smart [terminals] that all the different frequencies can actually go there and can be processed,” she said.
While AI is a powerful tool that can make different aspects of satellites more efficient, industry still has challenges to overcome and lessons to learn.
Tricha said developing satellite-specific open AI models is easier said than done.
“People think AI is kind of like electricity. I’m just going to plug it in, and AI is just going to ooze out the other side. But it’s not, it’s really not,” he said. “It’s a model that requires really understanding your information, understanding your data, cleansing your data, because data is garbage for the most part. What you get is not really what you can use to train, and there’s a lack of talent” as well.
Industry also has to take into consideration the number of older satellites in space and figure out how those fit into the larger picture of AI-enhanced systems, he added.
“We have to intersect this future,” Tricha said. “Just like we talked about multi-orbit strategies and high throughput satellites for a very long time, we are talking about AI-defined or AI-driven satellite communication, but it’s a little bit further out,” he said.
However, “you will start to see early deployments very [soon,] especially around intelligent routing.”
While incorporating AI into space operations comes with its challenges, industry wants the defense sector to know that it is nothing to be afraid of. On the contrary, it is a tool that should be used to enhance existing capabilities.
“AI is not going to take your job, but a person who knows AI is going to take your job, that’s for sure,” Tricha said. “These are learning systems. And by definition, as [we] invest more into learning, the [models] will just become better and better. … It’s still early stage, so this is an evolving technology. It evolves every week. It’s not going to go away.” ND
No comments:
Post a Comment