Lockheed Martin’s Melbourne-based STELaRLab looks modest from the outside, but it’s nurturing some exciting new thinking inside.
With the exception of a small student-built drone in its reception area, there is little to distinguish Lockheed Martin’s Melbourne-based STELaRLab — Science Technology Engineering Leadership and Research Laboratory — from any other office in the city.
As this is the defence giant’s first such facility outside the United States, it’s fair to expect some cutting-edge gear. But when asked, the centre’s director, Tony Lindsay, jokes the coffee machine is as advanced as it gets. So, what’s going on inside the rather generic walls of this $13 million research and development centre?
STELaRLab’s work is focused on autonomous robotics, hypersonics, quantum computing and data analytics. It officially opened in August 2017, and its current aim is to make better use of the vast amounts of information generated by various defence platforms.
Rather than housing expensive technology on site, it borrows what it needs from university and industry partners.
“We have access to a quantum computer based at the University of Southern California Information Sciences Institute,” said STELaRLab Principal Technologist Glenn Frankish. He added they might soon be accessing one at an Australian university.
“It’s a bit like a swimming pool, in that it’s great if your neighbour has one.”
Unlike Skunk Works, Lockheed Martin’s advanced aeronautics research centre in California, STELaRLab is a multidisciplinary corporate lab. Lindsay stresses that no weapons will be created as a result of its work.
“We have a connection across all Lockheed Martin laboratories,” explained the former defence scientist.
“However, because we’re corporate, we’re not allowed to do single-discipline things like aeronautics.”
Of STELaRLab’s 15 staff members, 13 are engineers of varying disciplines.
“We’ve got everything from high-performance computing engineers, hypersonics engineers, and people who’ve worked in millimetre-wave communications for the gaming industry,” Lindsay said.
Machines with intent
A key focus of STELaRLab is turning data into knowledge via C4ISR — Command, Control, Communications, Computers, Intelligence, Surveillance and Reconnaissance — solutions.
“The model for Australia is a small but very potent [defence] force,” Lindsay said.
“It comes down to how you maximise the effectiveness of that force. We’re looking at analytics and complex resource allocation problems. And, of course, once you start automating that, the machine learning comes in.”
Lindsay believes current machine learning has limitations in the defence sector.
“It is very good at understanding features and correlations in data without a human, but what it does with that data is told entirely by the human,” he said.
“How do you bring together all the information and the little signs that things are about to go wrong?”
This is where STELarLab’s innovation may prove most potent. It is developing ‘intentional machine’ programs that will allow them to contextually adapt to changing situations.
“Machine learning now is incredibly impressive, but it’s not actually the way people think,” Lindsay said.
“The machines have no way of adapting or contextualising, whereas an artificial intelligence, even in a constrained environment, will know what to do when it sees something different, even with just a limited number of rules.”
Werewolves versus villagers
STELaRLAb is aiming to inject some fun into this serious world of defence. It recently challenged a group of honours students to build an artificial intelligence that could play the game Werewolf better than humans.
A social deduction game that secretly divides players into teams of werewolves or villagers, the objective of the werewolves is to eat all the villagers before they are identified and killed off first.
“We scraped data from online games and did sentiment analysis,” Lindsay said.
“We built the cognition architecture to see if we could play better than a human.”
The humans won the day this time. “I think the cognition architectures at this point are fragile,” Lindsay said.
“They’re good when you tune them, but you’ve got to do a little bit of tuning.”
Researchers at STELaRLab will continue with this fine-tuning.
“We’re growing the lab, and we’re still very young and are only just ramping up the intentional machines program,” Lindsay said.
“The way to provide appropriate levels of security for the nation has to do with a trade-off between understanding intent and just mining everything because you think it might be worthwhile. There’s a balance between those two things. You have to be able to do that, but in a socially responsible way.”
This article was originally published as “Aiming for the stars” in the November 2018 edition of create.