IN DETAIL: Building for the future – LM AST on its R&D focus

In 2015 Lockheed Martin realised it need to more broadly engage international R&D opportunities. So it looked around the world for the site of its first non-US R&D centre and found it in Melbourne. Originally named the Lockheed Martin STELaRLab, the company’s Australian-based R&D centre is now called Lockheed Martin Advanced Systems and Technologies: it employs 65 staff, has nodes in four Australian cities and was a foundation partner in Adelaide University’s Australian Institute of Machine Learning (AIML), now a world leader in computer vision

Gregor Ferguson

One reason why Lockheed Martin chose Melbourne, says the company’s Advanced Systems and Technologies head Dr Tony Lindsay,  was the city’s research capacity and the country’s research quality. Melbourne is home to half a dozen universities – including the University of Melbourne which ranks 13^th in the world in the latest QS World University Rankings and is widely regarded as No 1 in Australia – and Australia has unrivalled depth of talent in the technology areas that are defining the future of national defence, especially autonomy, machine learning, hypersonics and space.

AST has a key role in funding and supporting the research of others, for example the joint Fire OPAL program with Curtin University in Perth, WA, which aims to map and find a path through the space debris and growing number of satellites in orbit, according to the LM AST GEOINT capability lead Dr Tim Payne.

“There is so much space debris in Earth’s atmosphere that in the future it may just become dangerous to leave Earth and we may get trapped in our own garbage,” he says. The number of possible catastrophic collisions between crewed or uncrewed spacecraft and pieces of orbiting junk or even just other spacecraft could grow almost exponentially as mankind continues to put growing numbers of satellites into orbit.

If AST has an equivalent, it’s Lockheed Martin’s Advanced Technology Laboratory (ATL).  ATL has a 95 year history of applied research and innovation spanning human systems, robotics and autonomy, spectrum operations, cyber operations, data analytics and advanced Command, Control and Understanding.  Like AST, ATL develops and transitions advanced concepts to Lockheed Martin’s four Business Areas, where they are matured to products that provide solutions across many aspects of National security.

The purpose of the LM AST laboratory, of course, is to commercialise the fruits of its applied R&D. It provides technology to all parts of the company, in both Australia and internationally, and has racked up a significant record of achievements in just eight years. It has developed and transitioned (and is still transitioning) capabilities for the RAAF’s Integrated Air and Missile Defence Command and Control, formerly the Joint Air Battle Management System (JABMS) being delivered under Project AIR6500; it has developed the Agile Shield counter-UxS battle manager; established strong and trusted Operations Analysis/simulation capabilities; and established new capabilities in sovereign high-speed flight analysis.

From the lab to the front line

While the ultimate measure of success is getting capability into frontline service, says Dr Lindsay, the lab moves technology from the lab to the final product carefully, he says, especially in areas such as AI and ML: “There are significant consequences in getting it wrong.”

Lockheed Martin spent the better part of US$1.6 billion last year on customer- and self-funded R&D globally. AST in Melbourne is a corporate asset and so competes for its share of that funding  with the company’s other R&D centres, including the Lighthouse and LAIC; it earns its money by delivering useable technology and insights that nobody else can – or at least not at the price.

The company won’t disclose what it spends each year on AST but cites Professor Emily Hilder, the former head of Defence’s Advanced Strategic Capabilities Accelerator (ASCA), who has been on the record as saying that Lockheed Martin is one of the major private sector contributors to defence R&D in Australia. The biggest single contribution to Australian Defence R&D comes from Defence’s own Science and Technology Group (DSTG) which spent around $677 million last year.

AST is working in areas where Australia has particular strengths – in some cases it leads the world. It has 65 staff – all Australians and all security cleared – organised into seven Research Teams covering Signals, Tracking, Fusion and Control; Mission Systems and Analytics; Machine Reasoning; Electromagnetic Systems; Hypersonics; Operational Analysis (OA); and field trials of Integrated Systems.

Between them these teams focus on Integrated Air & Missile Defence (think Project AIR6500); Multi-Domain Long-Range Precision Fires (think HiMARS and GMLRS and also hypersonics); Intelligence; Strategic Sensor Integration; Analytics and BMC2 for complex battlespaces; All-domain Counter-UxS Systems; AI/Machine reasoning; Operational Analysis (OA); Tracking and Sensor Fusion; and Advanced Sensing, including quantum sensing and processing of the signals. These are all technology priorities under either Defence’s 2024 innovation strategy, Accelerating Asymmetric Advantage, or AUKUS Pillar II.

Interestingly AST doesn’t use the almost-traditional NASA-derived Technology Readiness Level (TRL) scale to measure and present its developing technology. Instead, it uses a US Department of Defense budgetary scale which Dr Lindsay believes is a better measure of technology development for managers and financiers. Level 6.1 is the lowest: Basic research; 6.2 is Applied Research; 6.3 is Advanced Technology Development; 6.4 is Advanced Component Development and Prototypes; and 6.5 is System Development and Demonstration.

Beyond 6.5, which is where AST’s responsibility usually ends, is 6.6 – RDT&E Management Support; 6.7 is Operational Systems Development; and 6.8 is Software & Digital Technology Pilot Programs.

The majority of AST’s programs fall into the 6.2-6.5 region. However, one particular 6.5 technology is actually operational, so is close to TRL 9 under the old scale, while another, part of Project AIR6500, is currently in transition to Lockheed Martin’s Rotary and Mission Systems (RMS) DevSecOps (Development, Security and Operations) team.

 

Artificial Intelligence and Machine Learning

The LM AST approach to machine reasoning, where the lab was a founding partner in the University of Adelaide’s Australian Institute for Machine Learning (AIML), typifies the company’s approach. It specifies and funds specific R&D programs and aims to act as a bridge between the academics and the delivery of research impact, says Dr Leon Clark, AST’s Group Lead for Machine Reasoning. The AIML now employs around 200 staff and aims to deliver significant breakthroughs in all fields, civil as well as military, driven by nationally significant levels of investment. This is a multi-disciplinary eco-system in which Australia has become a highly capable international player.

The core competencies of the AIML are machine learning, trusted autonomy and human-machine collaboration. So far for AST it has delivered things like adaptive swarm control and trusted autonomy – essentially enabling swarms of autonomous vehicles to work together constructively to achieve a common goal; GEOINT – automatic target recognition using multiple sensors; high-level information fusion: and cognitive sensemaking via neuro-symbolic reasoning – using semantic abstractions to teach machines how to ‘think’.

‘Sensemaking’ is a pervasive issue. Most Artificial Intelligence and Machine Learning (AI/ML) systems are responsive, which is alright in a familiar environment, says AIML’s Director, Dr Simon Lucey. But warfare isn’t like that: it’s unstructured and non-linear and part of Lockheed Martin’s role is therefore to expose AI systems to real-world scenarios and the often-random nature of warfighting.

For Dr Lindsay decision-making in warfare is about intuition and an understanding of consequences: humans have intuition so can make contextual judgements based on their commander’s intent. AI systems don’t and have no sense of consequences, he says, so the potential for inappropriate or irrelevant decisions is immense. The emerging science of neuro-symbolics helps machines to ‘think’ more intuitively. Lockheed Martin is investing heavily in AI to assist its own internal operations, in addition to selectively investigating trusted AI approaches in  projects such as AIR6500, Australia’s Integrated Air and Missile Defence (IAMD) System.

Machine learning is going to be all-pervasive, going forward, says Tim Payne: the trick is not to do away with humans, but to find the right level of engagement, which again is a contextual thing.

The AIML juggles multiple challenges, says Clark: it must maintain currency with the state of the art globally; it must invest in the next generation of both technology and researchers; and it must empower academia to innovate, not just invent, and form effective relationships with customers, internal and external sources of funding and with research and commercialisation partners.

FireOPAL – rapid mapping

And the AST must find a research niche and not duplicate what other researchers are already doing, says Dr Lindsay.  For example, while lots of people are studying Space Domain Awareness (SDA) and mapping the Low Earth Orbit (LEO) region out to about 2,000km, the focus of the FireOPAL program with Curtin University and US companies is doing so quickly.

FireOPAL leveraged Curtin’s original camera-based Fireball program; about seven years ago Dr Greg Madsen, LM AST lab’s Engineering Lead for FireOPAL, decided this was the right technology on which to build a very fast satellite mapping and analytical tool.

Part of the technology at its core is still very simple: a $2,000 Digital SLR camera (you can get them from Ted’s Camera Store), along with a series of specialised algorithms developed by Lockheed Martin. The FireOPAL system works from sunset to sunrise, with the sun’s oblique rays illuminating targets in the sky. It is able to observe 97 per cent of all the objects in the space satellite and junk catalogue and has spotted targets at ranges of 65,000km.

The AST maintains sensor and computing resources at the company’s Uralla facility in NSW where detections are passed to a US partner company for further processing. This partnership has demonstrated a unique ability to execute persistent, wide-area space surveillance of reducing the time to detect, report and alert other space users of a new threat – it could be a new satellite or an unexpected move by an existing satellite – to just 110 seconds. The data is supplied to the ADF, US Space Force and civilian agencies. This is a vital capability for both military and civilian space users.

Atmospheric attenuation means cloud can obscure targets, which makes FireOPAL weather-dependent. But the system at Uralla can be reproduced anywhere there are communications and power, says the company, so using multiple sensors – from 5 up to ten cameras in different areas – can get around this problem.

Flight heritage in Hypersonics

A technology priority of both Defence and AUKUS Pillar II is hypersonics – flight at speeds greater than Mach 5. Australia is a world leader in this science thanks to decades of research by the University of Queensland and then by DSTG with a variety of US partners.

The problem is that hypersonics is still very much an experimental science. Research centres, government agencies like DSTG and companies large and small simply don’t have hypersonic test vehicles on which they can achieve ‘flight heritage’ for everything from new materials to guidance systems and sensors. So, there is a push under way to create a family of hypersonic test vehicles that can deliver both a high flight tempo and a high volume of experiments in order to drive down the cost of hypersonic flight by getting everything flight tested and certified quickly.

AST is working with the University of NSW under the Defence Trailblazer program to develop a modular high-speed flight testbed for experimentation and to mature hypersonics technology, according to Dr Scott Beinke, AST’s hypersonics lead. It has put its resources into developing a Common Front End (CFE) that can withstand the extreme pressures and temperatures anticipated in hypersonic flight, that is recoverable and that is agnostic about the means of hypersonic propulsion. All the propulsion system needs to do is go to an altitude of about 60km and a speed of Mach 7.

The CFE has two configurations, depending on the type of payload, and enables multiple experiments per flight, according to Dr Beinke. The AST research program is designed both to accelerate the maturation of Australian hypersonics technology and to train and develop a skilled Australian workforce and a skilled supply chain. This in turn would support both AUKUS and Defence’s technology needs as well as those of Defence’s Guided Weapons and Explosive Ordnance Enterprise (GWEO) – because in getting to hypersonic speeds and then slowing down again a research vehicle has to pass twice through missile-type speeds (around Mach 3), so the CFE could also be used to test GWEO-type technology insertions. And of course the hypersonic regime overlaps with the space regime so there is no reason why larger systems couldn’t be used to test some space hardware also.

Ultimately the ADF determines AST’s R&D program, says Dr Lindsay: you should always let the warfighter guide the R&D, he maintains. But the Lab also tests those warfighters, he adds: the warfighters know their jobs intimately but they don’t always know what advanced technology can deliver in addition. Technology, or even simply a new way of thinking, can unlock new capabilities, and those capabilities are AST’s goal.

To use a different metaphor, AST’s focus is on getting from A to B in the fastest time possible – and that could mean developing a new means of transport rather than simply breeding a faster horse, to mis-quote Henry Ford. So, AST’s mission is to disrupt, where necessary, on its way to delivering the ADF with an affordable, technology based capability edge.