AFRL researcher impacts community with runtime assurance
Editor’s note: To view the entire 38-page spread, a subscription is required and can be found at https://ieeexplore.ieee.org/document/10081233.
WRIGHT-PATTERSON AIR FORCE BASE, Ohio (AFRL) – Dr. Kerianne Hobbs, with the Air Force Research Laboratory, or AFRL, was the lead author of a 38-page spread in the Institute of Electrical and Electronics Engineers, or IEEE, Control Systems Magazine, titled Runtime assurance for safety-critical systems: An introduction to safety filtering approaches for complex control systems.
Hobbs worked on several autonomy programs at AFRL since 2011 and has been exposed to working with various systems, such as automated air refueling; Auto GCAS; and Automatic Air Collision Avoidance System, or Auto ACAS, which led to her current position as the safe autonomy and space lead with the Autonomy Capability Team, or ACT3, for the Sensors Directorate at AFRL.
Runtime assurance, or RTA, is used in various systems, most notably the Automatic Ground Collision Avoidance System, or Auto GCAS, which prevents aircraft from ground collision by automatically controlling the aircraft, saving lives and aircraft.
She started her current team in 2020 with the intent to develop ways to assure the safety of artificial intelligence, or AI, based autonomy of air and space systems. Having worked in the autonomous controls branch from 2015-2019, Hobbs saw how very complex systems are verified and validated.
“To me, an AI-based autonomous system was pretty much the ultimate challenge,” she said.
In a fast-paced field where there is a lot of new research, Hobbs understood the importance of keeping up with the latest technology.
“Some approaches, like the Auto GCAS, use switching-base runtime assurance, where we have some kind of backup control to switch to,” Hobbs said. “There’s a whole new field in optimization techniques where you consider what the primary controller would do — like the human or an AI based controller. You try to find something that’s as close to what it wants to do as possible, while guaranteeing multiple safety constraints in an optimal way.”
Looking at the big picture, Hobbs reviewed what was done in the past, and state-of-the-art technologies in the field, when considering designs for the systems.
“I did my Ph.D. dissertation on runtime assurance, and most of my work at that time was focused on the switching-base runtime assurance,” she added. “And we’re starting to see more advancements in the optimization-based runtime assurance for different scenarios.”
Hobbs said she dove into the area of runtime assurance, where rather than trying to prove an AI system is going to be completely safe offline, she looked at a way to wrap AI control systems in an RTA controller. This technology may allow airplanes and spacecraft to be monitored while assessing what the AI would do next.
“And if the AI is going to be unsafe, the RTA is able to modify or substitute a different signal that will guarantee safety,” Hobbs said. “So, we can provide a higher level of assurance or proof of safety by having this safety wrapper.”
The idea of the 38-page spread in the IEEE Control Systems Magazine was to capture the current state of the theory of RTA and answer questions such as what are all the different approaches? How do we categorize them? And what are some of the different spin-offs of related concepts? Hobbs added.
Dr. John Schierman, senior research aerospace engineer at AFRL’s Aerospace Systems Directorate, met Hobbs when Schierman was a contractor and continued collaborating with her as he joined AFRL.
“I was a contractor for about 20 years, and most of my research was on runtime assurance,” Schierman said.
Even when Hobbs moved to the Sensors Directorate, Schierman and Hobbs continued to collaborate on runtime assurance.
Hobbs’ career is just taking off, and Schierman said she has already made a big name for herself in the runtime assurance area.
“IEEE Control Systems Magazine is a very prestigious magazine,” Schierman said. “It’s a very prestigious thing for her to present her material.”
The magazine is known for giving authors a number of pages to fully cover and review a lot of the technical work, Schierman added.
“I think it highlights the expertise that AFRL has and that we’re spearheading in this area,” Schierman said. “And for her to be the lead author and to show that she is from AFRL, I think that’s very prestigious in our research community.”
The paper offered a tutorial and examples, focused on safety critical systems and summarized current happenings in the field.
“It’s definitely a big deal to have the cover article,” Hobbs said. “One of the reasons it’s a big deal is that the journals use something called an ‘impact factor’ to measure the impact of a publication or journal.”
Hobbs said an impact between four and six is significant for aerospace, and this publication has an impact factor of nearly six.
While Hobbs is already happy with the article’s reach, the initial and future goal is to compile a textbook on RTA. Hobbs and the co-authors of the journal article knew writing a textbook would take many years, so they began smaller, capturing what would be an outline for a future textbook, Hobbs said.
The article also aimed to answer additional questions, such as “What does it mean to do runtime assurance? What does the field look like right now, as a snapshot? What are the different elements of it? How do we categorize it?” Hobbs said.
While still working on her Ph.D., Hobbs worked on the article for about two years before it went through a peer review process.
“It’s one of the most extensive peer review processes I’ve ever been through,” she added.
Many times, there are one or two reviews with various scientists who provide feedback; however, this peer-review process offered extensive technical feedback with three or four reviews.
“It took about two years to get through the peer review process and publish,” she added. “In that process, it just got better, it got stronger.”
Before the article was fully released, a pre-release was published.
“We put up a pre-release in 2021, called a preprint, on this ArXiv website,” Hobbs said. “It already has 12 citations, which is a pretty big deal for a preprint article to have that many citations.”
This is especially important as researchers are looking at the use of deep learning systems while assuring safety, “especially when people want to put deep learning systems on robots,” she said.
Hobbs felt as though this article was a turning point, and she was able to accomplish influential work.
“For me, I think I was just very proud to actually go through this process,” she said. “I can do good work that matters, and that people care about. Not only in AFRL, because it has a huge impact potentially on Air Force and Space Force missions, but the larger community.”
Lab to warfighter
Mark L. Mote, cofounder of Pytheia, a startup that works in AI, and the second author of the magazine article, met Hobbs in grad school in a lab that focuses on safety for robots and autonomous vehicles.
With similar interests around applying these problems to aerospace systems, Mote and Hobbs collaborated on RTA with the concept of “doing safety better,” Mote said.
“It’s really just getting the word out and making RTA more concrete as a field and as an approach to safety,” Mote said. “A lot of people don’t really know how to go about these safety problems.”
There are many ways to approach the problem, but Mote said this is a fundamentally different approach.
“We can take any system that’s potentially unsafe, and we just wrap it in something that guarantees safety, and we have to know nothing about what’s actually at the core of the performance driven algorithm.”
Because one mission of AFRL is creating technology in the lab that the warfighter can use, Mote said RTA is a practical solution.
“If a pilot passes out and the jet begins to fall toward the ground, without the RTA systems, unfortunately, the plane hits the ground,” Mote added. “With what we’re doing, we have the potential to have a last-minute safety mechanism that comes in and pulls the plane up.”
The Auto GCAS is an example of this runtime assurance that has saved 13 lives and 12 aircraft. The technology is an example of an AFRL top priority, “from lab to warfighter.”
The Auto GCAS is a manned airplane where the primary controllers are human, Hobbs said.
“As we’re looking at potentially developing more and more complex autonomous technologies — having additional systems that can provide safety guarantees — means that even when the autonomy messes up, the aircraft can continue to be used another day,” Hobbs said. “I think they are going to be really important.”
RTA technology works during operations or a flight test by monitoring the primary control of the aircraft and its location, Hobbs said. Simultaneously, it can check for safety and intervene with an automatic function that will take control of the aircraft or the spacecraft and bring it to a safe condition before returning control to the pilot.
“I think the biggest thing that this helps with is the development and test of new autonomous control systems,” Hobbs said. “A control system for an aircraft basically takes whatever the input is from a human sometimes, or it could be your computer based, telling the aircraft or spacecraft where to go. And then it actually does that using different actuators on the airplane.”
The control system figures out where to go and how to change by interacting with the world, Hobbs added.
The technology to verify novel autonomous systems is typically a decade behind the initial technology; Hobbs said she hopes the RTA technologies will provide a stopgap.
At the Test Pilot School at Edwards Air Force Base, California, there are two test aircraft: the Calspan Learjet and X62A VISTA aircraft, which both have a form of RTA onboard and are being used to test novel autonomy algorithms, Hobbs said.
They have an envelope protection monitor RTA, which takes what the aircraft is capable of, such as the maximum velocities, and shrinks it.
“Then they say, here’s the sandbox you’re allowed to fly in, and we can test these different technologies,” Hobbs said. “But if you ever go outside of that, we’re going to switch to a human controlling it, so they can recover the airplane.”
Hobbs hopes the research expands it to include geofencing, where the autonomy controller will be forced to stay inside a specific map area.
“One of the biggest challenges right now is trying to combine multiples of these different types of safety systems together,” Hobbs said. “A lot of our research and looking at optimization-based techniques, as well as trying to figure out what is the right way to combine these runtimes assurance algorithms. We can provide some guarantees of multiple safety constraints. It’s never going to violate these constraints because these test assets are so valuable, we don’t want to lose them during the test process.”
Mote said it is an exciting time to be in a STEM career, in particular, with the advancements seen in AI.
“For this type of research, someone’s going to have to figure out how to make these very powerful things safe and control them,” Mote added. “I think the field is changing faster than ever before. We’re seeing some crazy exponential growth.”
Because of her successes, Hobbs encouraged others to follow their passion in science, technology, engineering and math, or STEM, career fields.
“I think people are motivated by two core emotions, love and fear,” Hobbs said. “A lot of times people, especially women or minorities, who maybe love science and they find it really interesting, may be afraid that they can’t do it or that it’s going to be hard.”
But Hobbs said following the career passion and putting in the hard work can help people conquer those fears.
“It’s extremely rewarding,” she said. “So, it’s very hard work, but I think most of the things that are hard work come with that reward or satisfaction at the end. If it’s something that you do love, find that courage to squash the fear and go for it.”
As the lead in this area, she’s collaborated with several graduate and undergraduate students, and Schierman said it is an important part of her job — mentoring students.
And while she may be the only female author on the 38-page spread, Schierman said she has worked with several female graduate students and helped find funding for them to work at AFRL as summer interns.
“Some of her students have graduated and continue to work in this area and collaborate with her and her team,” Schierman said.
AFRL also offers STEM employees the chance to become technical experts in their field, and Hobbs said she appreciates AFRL for helping her along the path to becoming a technical expert.
“There’s a variety of different jobs you can do, from program management, to engineering on a bunch of different programs, to going into this deeper technical expertise route,” she said.
She said becoming a technical expert in any career field is mutually beneficial. When there are questions in specific areas from the Department of the Air Force or the Pentagon, the technical experts at AFRL are sought out.
She said AFRL also monitors low technology readiness level capabilities, and the work from other industry partners, to see when they are ready for the Department of the Air Force.
“It’s this really cool place to be at the seams of the advanced engineering and the fundamental research and looking at what the best technologies are to invest in for the future of aerospace,” Hobbs added.
The Air Force Research Laboratory is the primary scientific research and development center for the Department of the Air Force. AFRL plays an integral role in leading the discovery, development and integration of affordable warfighting technologies for our air, space and cyberspace force. With a workforce of more than 11,500 across nine technology areas and 40 other operations across the globe, AFRL provides a diverse portfolio of science and technology ranging from fundamental to advanced research and technology development. For more information, visit www.afresearchlab.com.