Artificial Intelligence (AI) is ushering in a new age of information and the applications for the satellite industry are vast. AI can generate efficiencies throughout the satellite life cycle, from manufacturing to operations, which may be key as constellations will vastly increase the number of satellites in space in the coming years. And as Earth Observation (EO) satellites continue to capture higher levels of resolution, AI can transform how data is processed both in space, and on Earth, increasing the speed at which insights can be delivered to customers. In this round-up, industry leaders from companies like Airbus, Relativity Space, and Hypergiant share how they are using Artificial Intelligence and Machine Learning (ML) to increase space capabilities and make a difference in the satellite industry.
Currently, we’re implementing ML-based orbit prediction algorithms which we are using to predict the position of satellites and debris in Low-Earth Orbit (LEO) with greater accuracy than other publicly available resources. We’re also implementing object detection, classification, and localization algorithms on the satellites themselves via Convolutional Neural Networks. This gives those satellites the capability to identify and respond to objects they observe with on-board sensors without having to wait for communication from ground stations. Within those ground stations, we’re ingesting telemetry data from the satellites and using that data to classify and predict anomalies using a variety of supervised and unsupervised learning techniques.
Finally, one of the major facets of our AI and ML work is with assisting human operators managing large fleets of satellites. In this realm, we are developing algorithms to automatically monitor, maintain, and task satellites and ground systems; to automatically respond to stimulus from a fusion of sensors (both on the satellite and from the world at-large); and to efficiency route uplinks and downlinks through a mesh network of ground systems and satellite nodes. The overarching goal here is to hand the tedious, rote, "low cognitive tasks" over to artificial systems while keeping our irreplaceable human operators focused on the rare but critical, “high cognitive tasks” associated with these constellations.
— Ben Lamm, CEO of Hypergiant Industries
At Relativity, we are laser focused on incorporating new technologies that allow us to solve problems never addressed before, automate aerospace manufacturing, and further revolutionize how rockets are built and flown. A key part of this has been realized in how we embrace Artificial Intelligence and Machine Learning in our manufacturing processes.
Our team has built the world’s largest metal 3D printers internally, and their tech stack wouldn’t be possible without our proprietary and patented AI. Layering our AI and ML software with intelligent robotics has allowed us to optimize every aspect of the rocket manufacturing process. In 2019, we were granted a patent for our advanced AI-powered sensors that provide real-time adaptive control. This technology provides our team the ability to create customized mission specific solutions and rapidly turn them into reliable flight parts, while also reducing lead time and part count. At Relativity we recognize that part of solving problems never solved before means being audacious in creating new solutions and embracing new technologies.
— Brandon Pearce, Avionics and Integrated Software, Relativity
At LatConnect 60, we see a tremendous opportunity to use Artificial Intelligence onboard our satellites to enable near real-time collection and delivery of Earth Observation insights at scale. Our patented Machine Learning algorithms can be applied on-orbit to detect an anomaly and trigger a response. Autonomous responses could include tasking another imaging satellite to collect imagery at a particular timestamp or coordinate a co-collection activity of different data types at the same area of interest. Tasking commands and data collected would be relayed via inter-satellite communications links. Our on-board AI will be able to select the most optimal data link. With sufficient on-board processing hardware on each satellite, our algorithms can process, classify or fuse large volumes of data on-orbit to provide insights directly to end users when they need it. We are seeing significant interest in this capability from government and commercial clients. There will be a greater industry focus on delivering outcomes from smart satellites over the coming years.
— Venkat Pillay, CEO of LatConnect 60
Lockheed Martin is applying Artificial Intelligence across the product life cycle – from production to satellite operations. AI is increasing the speed at which satellites can be developed and tested. During key testing milestones like Thermal Vacuum (TVAC), we use an in-house AI system called T-TAURI, which combs through testing data to analyze anomalous results in a fraction of time – significantly decreasing schedule.
AI is also enhancing space capabilities through pathfinder nanosat missions like Lockheed Martin-developed Pony Express and La Jument. Both missions are testing SmartSat, a software-defined satellite platform which uses containerized apps that can be easily uploaded in-orbit. By training an algorithm on the ground, we can upload it to a SmartSat-enabled satellite and run it in real time. One app being tested will be SuperRes, an algorithm that can automatically enhance the quality of an image and enable exploitation and detection of imagery produced by lower-cost, lower-quality image sensors. SmartSat is also opening the door to heuristic pattern and anomaly detection, enabled by AI, which improves cybersecurity resilience on-board with automatic updates as new threats emerge.
Lockheed Martin is also working on ways to autonomously command constellations of satellites of all sizes. As increasing numbers launch, satellites will need to autonomously make trajectory changes like slew maneuvers, a process that is both time and processor-intense for operators. Compass ML is moving those calculations to the edge so vehicles can plan their next maneuvers with or without assistance from the ground or respond to tips and threats.
— Linda Foster, director of Innovation at Lockheed Martin Space Mission Solutions
With fiber connectivity and consumer demand for Internet in remote areas increasing, satellite can assist in filling the demand in underserved areas. Indeed, satellite represents the foundation on which the entire communications network depends as it is extremely versatile and can reach areas where no cable, fiber, or mobile network will ever be available.
Artificial Intelligence has the potential to help satellite evolve. In a world reliant on constant connectivity, a well-managed network of virtualized teleports and robust redundancy switching are vital components of a reliable service that can be better managed using AI.
Satellite operators need to evolve, from the functional mindset of providing capacity, to the strategic mindset of delivering a service. By virtualizing networks and utilizing the huge amount of data available, machine learning can eventually automate tasks and support operations. Gathering data from base stations significantly improves interference detection, and information from ticketing systems helps to predict potential interference. Feeding an AI-enabled collision avoidance system with telemetry data, mitigates one of the biggest operational risk factors in space. But we need to realize as well that this takes time and effort to develop – as we want it to be at minimum as failsafe as our traditional task handling.
The satellite sector must adapt and focus its resources on building a premium service. This will not only require data but also collaboration between individual operators, in order to deliver a secure future for everyone.
— Helen Weedon, managing director of Satcoms Innovation Group
Traditionally, data processing and exploitation occurs through ground systems when satellites are overhead to download the data. That takes time, which we may not have, particularly with constellations that have hundreds of satellites’ data to process. At Raytheon Intelligence & Space, we’re working on advanced on-board processing using space-qualified signal processors capable of hosting powerful AI and ML applications, where the satellite becomes the data collector, exploiter and disseminator – the brain and the nervous system. That will enable satellites to deliver actionable intelligence directly to the right person at the right time.
We’ve also developed advanced software AI and ML algorithms to perform mission-specific space-based battle management, command, control, and communications applications. When you’re on the front lines, in any domain, time is of the essence. Our AI and ML algorithms enable machine-speed processing for high volumes of data generated from proliferated LEO constellations of sensors.
— Jason Kim, business development executive of Space & C2 Systems at Raytheon Intelligence & Space
Airbus has been using AI for the past few years to improve the quality of the satellite imagery it delivers to customers. This started by working with Google’s open-source Tensor Flow for automatic cloud detection, removing manual checks before image delivery. This has now evolved into automatic change detection of objects such as cars, boats or planes – which is part of our Ocean Finder and One Atlas services. We have been successful in this because we have billions of square kilometers of imagery dating back to 1986, coupled with annotations such as cloud masks made by our experienced domain engineers. We have useful data to feed Machine Learning platforms and the more data from which machines can learn, the more effective the result.
The next steps are to apply AI to stacks of temporal images to monitor Earth but also make AI analysis of imagery onboard the satellite so that image requests can be automatically reprogrammed gaining precious time to generate useful insights. Airbus not only offers these capabilities for its own service products but also onboard the satellites it makes.
Moreover, we are putting AI to use to help us monitor the health of our satellites in-orbit to make future satellites better. We pool all telemetry into a data lake which our engineers then use to create algorithms relating to all stages of a spacecraft’s lifecycle. Last but not least, Airbus sees AI enabling ground segment automation, key to the efficient management of future mega satellite constellations.
— Jean-Marc Nasr, head of Airbus Space Systems
It’s not a new phenomenon to apply Artificial Intelligence in the space industry. But currently, widespread solutions mostly use ground-based equipment. Typically, satellites collect large quantities of business and scientific data and downlink them to the ground, where processing and analysis occur. In many cases, this is insufficient — downlink overload deriving from data volume requires relief, and the need for real-time information demands immediate processing. Not only precision agriculture or natural disaster damage mitigation but also telecommunications, docking operation, asteroid mining, and other autonomous satellite operations could make use of this capacity.
Hungarian C3S LLC. is developing a solution that integrates autonomous vehicle tech company AImotive’s neural network hardware acceleration technology into its space electronics platform to enable high performance AI capabilities in small, power-constrained satellites. The purpose is to get data processed onboard by using Artificial Intelligence, which results in end users obtaining real-time and tailored information instead of data sets waiting to be processed. This method of enhancing capacity is adaptable for large-scale satellites, and as it is general-purpose hardware, it is also applicable for nanosatellite missions.
The demonstration project for this solution is an Earth Observation mission that supports precision agriculture. The onboard computer processes an increased amount of high ground resolution hyperspectral satellite data. The onboard processing of hyperspectral camera data provides farmers essential information on the crops’ biochemical and biophysical state that can bring an increase in volume in the harvesting phase and hence, may contribute to the optimization of the global food chain.
— Gyula Horváth, CEO and co-founder of C3S
At Hughes, we use Artificial Intelligence and Machine Learning to streamline network operations and improve the customer experience — from planning to installation to ongoing network optimization. In the planning phase, we developed an AI application to identify what data transports are available and aligned to our enterprise customer’s business needs at any of their locations. For residential and business satellite installations, another AI-driven app that we developed automatically reviews completed installations and identifies any issues that need to be addressed to help get our customers online faster.
Across our network, we apply AI for traffic management, triage and capacity planning, often implementing pre-emptive actions to shorten help desk response time and avoid issues before they happen. Hughes is the first managed services provider to deliver a self-healing WAN edge capability to enterprise customers — in use at more than 32,000 sites. This AIOps feature automatically predicts and preempts undesirable network behavior—preventing issues in 70% of cases and providing early diagnoses in the other instances.
And finally, in the defense arena, we are applying AI to the Flexible Modem Interface (FMI), part of the $2.2 million U.S. Space Force (USF) Space and Missile Systems Center (SMC) contract. Built into the FMI, an AI rules engine autonomously makes decisions based on the terminal’s state, operational environment, and relevant policies — enabling resiliency and reliability for military networks and the flexibility to change configurations in near real-time.
— Sharyn Nerenberg, senior director for Hughes
Artificial intelligence is elevating how the satellite industry can be used at scale. Today’s abundance of satellite imagery has huge potential to help companies, investors and governments make critical decisions, but the volume also makes it impossible to manually analyze the trends within each of those pixels. Applying AI helps make sense of this data deluge by automating analysis and even integrating it with other data. My company, Orbital Insight, develops geospatial analytics to reveal previously hidden trends about what’s happening on and to the Earth. We use AI to transform multiple sources of geospatial data — including satellite images, mobile location, connected cars and other Internet of Things (IoT) data — into objective answers about the state of supply chains, global commodities, geopolitical events, and demographics. The goal is more informed decision making.
AI vastly increases the power and applications of satellite imagery. For example, Unilever recently announced its partnership with Orbital Insight to track its palm oil supply chain down to the elusive farm first-mile and to prevent deforestation. After creating crop origin maps using geofencing and location data, our computer vision algorithms are applied to satellite imagery to monitor forest health in real-time and to watch for deforestation threats. During the pandemic, Orbital Insight has been tracking the effects of COVID-19 on the movement of people and goods globally to help organizations adapt.
— Dr. James Crawford, CEO of Orbital Insight