Space Threats: Tracking & Monitoring
With recent worldwide events, it has become clear that the Space Domain is becoming an integral part of the battlefield. This is not only for communications, but also counterspace activities.
Earlier this year, Russia launched their Luch (Olymp) 2 (“Luch”) satellite, which is being widely monitored by the space community to understand its intention.
I recently had the chance to speak to Audrey Schaffer, former White House Space Policy Official and Vice President of Strategy and Policy, Slingshot Aerospace on how their AI driven platform can help monitor and predict the movements with the Space Domain.
Q: Can you provide a brief overview of the Slingshot Aerospace platform
A: We are building what we call a digital space twin, which is a virtual representation of the space environment. It has a realistic physics based engine to model behavior in outer space and it ingests data from a variety of different sources. This includes publicly available data from the US government, proprietary data of ours from a global sensor network that we operate as well as data sources from some of the arrangements we have with our customers. We bring all of that data into one place and from our platform we are then able to build a number of different applications depending on what our customers are looking for. For example, if you think about modeling everything in the space environment all at once, that is very useful from a space traffic coordination perspective to be able to avoid collisions, detect close approaches and enable communication amongst our operators.
It is also useful for detecting behavior anomalies in the space environment. So for watching everything that is going on in outer space, we are looking at things like patterns of life, we can detect maneuvers that may or may not have been expected and then we can put satellites on various watchlists to follow up on their behavior.
You can also imagine that these tools are useful for space operators and space training. So we have a robust activity going on with the US Space Force where we are helping them train their next generation of space operators in the fundamentals of space operations as well as helping with some of their wargaming and exercise activities.
Q: The Luch (Olymp) 2 satellite launched back in March 2023, and these anomalies that you published happened in September 2023. What has been the history with the satellite up until now?
A: As mentioned, Luch (Olymp) 2 first launched in March of this year and it initially stayed in an orbit around 50/51 degrees east presumably to do some in-orbit checks, which is normal for many satellites. Then around May of this year, it drifted to a new position at around 9 degrees east. For context, when I say this, I mean 9 degrees east of the meridian, that’s how we lay down satellites that are in geo-synchronous orbit which is the orbit that more or less stays stationary as the earth turns, so they look at the same place on the earth. Luch 2 drifts into this new position and stayed there for around 5-6 months. Our story really begins on September 26th. Our AI driven object profiling engine detected a maneuver of the Luch satellite, think of it like a little burst of movement that begins the Luch drifting once again. It started drifting again westward from 9 degrees east and then on October 2nd, we detected another maneuver that it began to slow down. From those data points, we were able to predict where we thought the Luch was going to stop. So based on its history of lingering at its spot at 9 degrees east for around 6 months and then the maneuver, we then projected that it would stop around 3 degrees east where there is a cluster of satellites and low and behold, on October 5th, it slowed to a stop really right where we predicted it was going to.
That’s our story on how our AI driven platform was able to detect the activity but also predict what was going to be the final conclusion.
Q: Would an operator normally announce if they were making a maneuver?
A: There is no rule that says that you have to announce every time that you are making a maneuver and it is not a global norm at the moment, but typically satellites in Geosynchronous orbit stay in the same location. Think about why you might want to be in geosynchronous orbit in the first place. It’s because you can basically stay looking down at the same part of the earth without needing to do maneuvers or spending a lot of fuel and just do minor station keeping adjustments. That truly is the intrinsic value of being in this orbit and is useful for a variety of applications whether that is communications, or detection of launches on the surface of the earth. You are staring at the same piece of the planet all the time. So while it is not unheard of for satellites to change their position in GEO, it is also not something that happens routinely. That’s why we were talking about our algorithms, we weren’t looking at the Luch 2 satellite, we weren’t looking for this type of behavior. We just told our algorithms to tell us if something interesting happens. In this case a satellite changing its position in the GEO belt is something interesting as that is not typical of a normal pattern of life for a satellite in this orbit.
Q: What issues are presented by the Luch (Olymp) 2 satellite positioning itself so close to another operators satellite?
A: It really depends on how close it gets to the other object. It if gets really close, it could present a spaceflight safety concern but even if it is not quite that close, it could present a security concern. Like we said, it is not necessarily normal for a satellite to reposition itself into a new slot, as these slots tend to be allocated internationally by the International Telecommunications Union (ITU). It could also make the operator of the other satellite have to carry out evasive maneuvers as they cannot predict what the Luch craft is doing.
Q: There was an original satellite, Luch 1 launched back in 2014, does the Luch (Olymp) 2 appear to be displaying similar behavior?
A: There are similarities between the two. If you look at the pattern of life of Luch 1 and then compare it to Luch 2, it is very similar in that so far, both hang out near a cluster of satellites for 6 months to a year and then presumably complete their mission there and then drift to a new location near a new cluster of satellites and repeat.
Q: So at the moment it has found a location, it is sitting there for the time being, we could predict it is probably going to be there for a couple of months and then move on.
A: It’s not like our data is predicting that it is going to stay there for 6 months, but if you look at the pattern of life of what it did previously and what the previous Luch did, you might expect that it stays in its current location for another 6 months or so.
Q: So when it does start to move again, you can use that data to predict and to provide that to other satellite operators so they can be prepared?
A: Exactly and the interesting thing that I found about our maneuver detection algorithms, if you look historically back over a maneuver that occurred, it is easy to see that this took place, you can see the satellite begin to drift and can see it maneuver and then slow down. That is easy to look at historically, but in the moment, in real time, it is often difficult to discern whether a maneuver actually happened or whether a tracking network of telescopes and radars just lost the object for a period of time or misplaced it. This is actually a fairly common thing for the publicly available data. So, in the moment it could actually be that you are not sure when you see a satellite appear in a place that is different to where you expect it to be. The interesting thing for us is our algorithm detects the maneuver driven by our object profiling and historical data analysis. We believed it was a maneuver and not just a mis-tag / cross-tag and then we confirmed that it had taken place by tasking our own global sensor network, which we control, to take some new observations of the object to confirm it was moving where we thought it was.
Q: You mentioned your global sensor network, could you provide some more details on how this is made up?
A: We have a couple of different types of sensors. We have sensors that are called staring sensors, meaning they look at the sky all the time at night and they are able to pick up essentially anything they can see in their field of view. Then we have other types of telescopes that are tracking. They lock onto a single object and are able to follow it as it moves across the sky and those are taskable. We tell it to look at a certain object and get some observations, versus the staring ones that just see everything that is passing by. So taken collectively, our sensors are able to see all the way from Low earth orbit up to and beyond geosynchronous orbit that we have been talking about and into what we call cislunar space, which is the area between the earth and the moon. So we are able to look at all of these regions and because we have both of these types of telescopes, we are able to take observations both at night, which is normally when you think of a telescope, but also during the day which gives us an impressive ability to track objects under a variety of different conditions.
Q: Could you walk through the image that you release alongside this story?
A: If you look at the picture, we are going to read the chart from right to left. Along the X Axis is longitude. So it starts at 10 degrees east on the right hand side and the left hand side is 2 degrees east. Think of this as basically a map of geosynchronous orbit laid down on top of the earth. Each one of the vertical lines that are labeled ‘Sat1’ ‘Sat 2’ ‘Sat 3’ etc. are the positions of individual satellites and like we talked about before, these individual satellites are staying relatively in the same position. You see little moves here and there, that is just then doing their normal station keeping operations, which while there is a little bit of variability, it is more or less staying in the same position. This is the kind of behavior that we would expect to see out on a geosynchronous orbit. The Y Axis of the chart is time and it starts at the top and goes down to the bottom. So we start in the top right hand corner of the image and the blue line that runs diagonally down through the image, that is the depiction of the Luch 2 satellites path. It starts around satellites 1 & 2 and that is where it had been hanging out from May until September this year at around 9 degrees east. You can see the orange box that says maneuver detected. That was the first maneuver that we detected from the data that was available and we confirmed it with our sensors. Then you can see the Luch’s blue line as it starts to move. This is it drifting, so as we go forward in time which is going down the chart, it also starts to move left from the east further west. You can see it drifting by satellites 3-8 and then around satellite 9 is where we see it begins to slow down. It is based off this that we can make our prediction that we think that it is going to come to a stop around satellites 10/11/12 because it’s rate of drift / movement has begun to slow and we see this cluster of satellites, which based off its previous activities, we think might be interesting to the Luch. So in fact In the end it ended up where we predicted. This picture was drawn before it stopped but shows our predictions on where it would end up.
Q: So right now we are in-between the two boxes in the bottom left hand corner of the image?
A: It actually stopped near satellite 10 so it never got far enough to get close enough to those after it started to slow down but we weren’t sure where it was going to stop. So this is where it would have ended up if it continued. But it stopped in the vicinity of these 3 satellites and that was our initial prediction all along.
Q: Can you provide more details on what your previous position involved?
A: I worked at the national security council at the white house, which is the body that brings together all of the parts of the United States government to develop national security policy for the country and to react in real time to global events. My portfolio there was everything having to do with space. So that was the longer term look at what the US policy should be for a number of topics in regards to space; whether that be international cooperation or thinking about the future of military and intelligence space activities. I was responsible for this along with the day to day management of crises and other incidents that would affect America’s security or foreign policy interests. For example I spent quite a bit of time dealing with the ramifications of the Ukraine war with regard to the use of space both within the conflict and the reverberations it had for our partnership with Russia for example on the ISS (International Space Station) and figuring out how we were going to thread that needle. I really served at the top of our government system orchestrating the day to day space policy making activity.
Q: What made you jump to the private sector?
A: There is just so much exciting work happening in the private sector, not just in the US but around the world. Companies like Slingshot Aerospace are doing things that were historically only done by governments. 5-10 years ago you wouldn’t have seen the private sector do what they can today. So for me it was a great opportunity to not only do something different but to also drive innovation in an area that I am really passionate about.
In addition to speaking to Audrey Schaffer, I also had the chance to hear from Clarice Reid, one of Slingshot Aerospace’s Senior Data Scientists, on how their machine learning and AI is pivotal to make this analysis possible.
How the Slingshot Aerospace Platform Predicts Targets of Hopscotching Satellites
The broad vision is that we have a variety of data sources that we ingest and infuse, and then stacked on that we have a number of insights, analytics and algorithms. Most of them are ML (machine learning) or deep learning based. We do have a few that are based on subject matter expertise. We really try to marry physics and astrodynamics expertise with ML and big data. All of these different insights run on the data to give us a 24/7 view of what is happening in GEO. A few also run on LEO in a reduced set.
We have maneuver detection, which asks the questions "Has someone maneuvered?", "Why are they maneuvering?", and "Where is the satellite going?" We also have satellite object profiling, which is both "What are the typical behaviors exhibited by this satellite?" and "Given a global view of satellites, what do we think the mission is right now?Is it active, is it station keeping, has it fallen into disrepair, are they trying to move it out of orbit?" A number of smaller flags and analytics are used to get a detailed view of what is going on in GEO. These can be bundled together and taken over a complete history of orbit, and then fed into an algorithm that decides what is typical behavior for GEO and what is atypical for GEO orbit.
The object profiling we run is able to detect the difference between an expected maneuver, like station keeping, and an unexpected one, like the start of a drift. This helps the algorithm and subject matter experts to decide whether something unusual is occuring, and whether anyone needs to be alerted.
How Slingshot Aerospace AI and ML is State of the Art
We are covering the spectrum on what is happening right now in AI, especially with deep learning. We started with the easier algorithms, using , basic astro and simple ML, and while those were in development, we had a team working on some of the bigger AI, like what's been in the news.
We have essentially been trying to apply all of these recent advancements to the space domain with a good amount of success. We have some really cool stuff happening in terms of object profiling, especially with some of the larger data sets. Near-earth orbits have gotten crowded and complicated in the past few years, and so it is extremely useful both for profiling constellations and satellites, and also predicting and simulating action up in space.
Over time, our predictions will be more accurate, as models train and as we get more data. Obviously the Luch (Olymp) 2 launched just this year, but the longer a satellite is in orbit, the more we have in terms of sampling its personal history and so understanding of behavior should improve as time goes on. In the meantime, we have data going back to the 1960s. While we can fine tune on individual action, the algorithms we have out now were trained using the full history of historical data.