Harnessing AI for Intelligent Sensors
The United Kingdom is advancing sensor, spectrum capabilities and autonomous systems for so-called Cognitive Intelligent Sensing (CoInS) technologies, for the U.K. Ministry of Defence and the U.K. Defence Science and Technology Laboratory (DSTL)—akin to the United States’ Defense Advanced Research Project Agency.
Several companies are conducting research and mounting partnerships to further capabilities to detect, track, sense in the spectrum, for drone operations and defense of other adversarial systems.
Under one such collaboration announced in May, Leonardo UK is partnering with artificial intelligence (AI) company Faculty.AI to bring lab-based or other emerging capabilities into CoInS and electronic warfare.
The partnership is Leonardo UK’s first under the country’s Collaboration Partner Programme, which enables innovators and companies of different sizes to advance joint products and services to market. The two companies previously worked together under specific contracted research projects in autonomy and electronic warfare, Leonardo reported.
The activity represents the flood of AI, machine learning and other autonomous systems into intelligence, surveillance and reconnaissance (ISR) research and development, with more and more embedded technologies going into ISR products and services.
The various research efforts are meant to accelerate the pace at which AI can be applied in the military.
Leonardo UK was already ramping up its AI capacity in general, given the demand growth for AI in defense and security industries, the company noted.
“The first joint projects will focus on the application of AI into CoInS and the technology will allow sensors to self-orientate themselves without the need for humans to operate them remotely,” Leonardo UK said in the report. “Leonardo and Faculty AI will also be looking at how AI can boost the capabilities of electronic warfare payloads and countermeasures for combat aircraft.”
The research will involve Leonardo UK’s BriteCloud decoy and the BriteStorm jamming system, with a focus on “practical outcomes” and leveraging AI for performance improvements, the company said.
“We will be looking for opportunities where Leonardo’s expertise in defence electronics sensors and integration, military rotorcraft and cyber security can incorporate the AI expertise of Faculty to deliver something of tangible benefit to our customers,” said Simon Harwood, capability director, Leonardo UK, in a statement.
Faculty.ai, meanwhile, began as a startup that placed academic fellows with companies to bridge the personnel gap between researchers in engineering, math, science doctorate and post-doctorate levels, explained Ross Adams, customer director for defence, Faculty.ai, in an interview with SIGNAL Media.
A mathematician by trade specializing in data science, data exploitation and delivery of artificial intelligence-related solutions, Adams sees the company further evolving as the AI renaissance continues to expand over the next several years and as they expand their offering of AI services and products. In addition to personnel placement, the company is specializing in creating bespoke AI and machine learning products for the military as well as customers in other sectors—energy, finance, insurance, retail, health and life sciences. Defense is their largest sector.
Faculty.ai also has a center of excellence for sensors, and when employees are not working on director and client projects, they collaborate with other researchers to explore what is coming out of universities. They examine what can be applied to sensor challenges—even if a technique isn’t initially used for radio frequency classification, Adams said.
One of the first military efforts of Faculty.ai was to help the U.K.’s new Defence AI Center in setting up a number of initiatives and exploring concept delivery considerations. The company has also worked with the DSTL, focusing on computer vision, autonomous software for drone deployment or signal classification, decision intelligence, cognitive sensing and predictive analytics, Adams shared.
The latest partnership work with Leonardo UK involves electronic warfare research, as Leonardo UK produces aircraft-based radar solutions to support the U.K. military and other global efforts. They are deploying Faculty AI’s technology and leveraging personnel expertise to improve internal processes and operations at nine Leonard UK sites.
“They are constantly looking at novel ways to try and improve their radars, so a lot of the research goes into that,” Adams said about partner Leonardo UK. “They are also looking at ways to counter drones, provide counter [unmanned aerial vehicle] security or find ways to deploy things onto a slightly smaller kit.”
Also, Faculty.ai and Leonardo UK are sponsoring several faculty fellowships, placing graduating master’s- and doctoral-level students in short-term industry placements.
“These will deliver impactful AI-based projects within Leonardo programmes, backed by support and training from Faculty AI’s team of experts,” Leonardo UK stated.
As part of its AI work, Faculty.ai is harnessing rapid research methods, including retrainable machine learning methods, and bringing in “SWAT team” researchers to enhance Leonardo UK’s internal research processes.
“We have data scientists with slightly different, complementary skill sets that look at data and use a variety of different methods coming out of universities to just try and spot patterns,” Adams shared. “And if there was an electronic signature, for instance, coming off a drone or a radar that we weren’t spotting, we developed methods that could be retrained overnight, so that the next day the product would be able to spot it.”
By using modern data science methods from academia, Faculty.ai researchers are leveraging one large language model approach known as a transformer. Originally used for words in a language, to understand the context of different words and sentences—essentially the techniques underpinning ChatGPT—the approach, when applied to spectrum, could be fruitful in the future.
“When it comes to radars, signatures, signals and pulses, you get these ‘pulse words,’ and they do look a bit like words,” Adams clarified. “So, we explored whether we could use transformer-based techniques to try and understand some of those signals, with some success, but the maturity of the research is not quite there yet from our perspective.”
In centering on retrainable models, the researchers are definitely prioritizing real-world data.
“The way that we procure that data and use it for development and testing is kind of crucial,” he shared. “We have been able to get involved with U.S. SOCOM [Special Operations Command] trials, for instance. We have flown the team out and collected a load of signal data from different emitters in a crowded space. On the first day, we were able to identify some emitters, but not all of them, and we were able to spot that we had some weird stuff going in the data that we weren’t tracking. But because we had that data, we were able to work on it overnight. [With] the supervised machine learning models and labeled data, if you’ve treated it in the right way, then you can retrain it.”
In addition, Faculty.ai is also developing a future aircraft-based capability around electronic warfare for the Royal Air Force, although Adams could not go into further detail.
And in a separate partnership with Talus, Faculty.ai is working with Talus’ U.K.-based AI business division, Cortex UK, to help build up their capability and business, assisting with recruiting, tooling and assured machine learning and other AI solutions. For one capability, Faculty.AI is helping build new features to detect anomalous behavior in aerospace and maritime operations.
“They have a large command and control type product, which would help military analysts receive a load of data from shipping or aerospace traffic, whatever it might be, and be able to [provide] situational awareness and make better decisions based on whether these are commercial entities or something more hostile.”
The solution provides a situational analyst’s overview, analytical features, visualizations, geographic overlays, temporal perspectives and tracking—whether ships around the North Sea, around the United Kingdom or airplane traffic.
And naturally, the “assured” part of any AI-based solution is vital, Adams continued.
“This is one of those common things that I get asked about,” he said. “And in the AI community, that is what we are all asking each other about. How can I deploy this and trust it?”
Part of that assurance involves extensive red team and safety testing.
“From a Faculty.AI perspective, we’re involved with some of the frontier AI models out there, and we are named on their system card as people who have red-teamed it and safety tested it,” Adams stated. “Can you trick it into doing something that it shouldn’t really be doing? Can you evade the safeguards? We have got red teams that are involved with all of that. It’s the same principle that we apply to any machine learning system, which is that everything needs to be tested adversarially.”
Comments