AI and machine learning are helping engineers take drone delivery to the next step
- May 16, 2019
A while ago, Dr Terry Martin predicted unmanned aerial vehicles (UAVs) were on the cusp of some very interesting developments, and a good match for his background in both aerospace and machine learning.
For Dr Terry Martin, Applied Research Lead at Nova Systems, UAVs were an attractive place to focus his efforts. They are, essentially, remote sensing platforms, and what they can do is growing rapidly.
“The futuristic vision is to extend that capability into so much more, where the data captured by the platform is turned into information and knowledge that supports decision-making and is distributed effectively,” Martin told create.
“And that’s a field of endeavour where I hoped to deploy my hybrid skills and experience. And that’s proven to be the case.”
He sees AI as something that will become more relevant as systems move from “human-in-the-loop” to “human-on-the-loop”, with machine learning, pattern recognition and image processing technologies seamlessly sorting and annotating what’s collected to support decision-making.
“That’s coming: it’s just the timeframe for how long it’ll take,” he said.
Martin is not alone in his interest in drones and is working at the frontier of a booming technology area that could radically reshape our lives over the next decade.
According to research firm Markets and Markets, the UAV sector’s value is growing at around 20 per cent annually and will be worth US$52.3 billion by 2025.
Porsche Consulting said it’s reasonable to expect one class of UAV within the next decade: passenger drones. The same firm estimates that market will be worth US$1 billion by 2025, then grow over 20 times that amount within a decade.
A massive challenge
As spectacular as a future of flying cars might seem, life-or-death drone missions – for example of emergency medicine or blood, search and rescue efforts, or a flight to thwart a terrorist – would also change life as we know it.
Before any of this happens, a massive degree of coordination is needed. Operators, telcos, regulators, data service providers and others must work together, establish standards and overcome numerous sets of challenges.
A key enabler for wide-scale and safe deployment of UAV and urban air mobility (UAM) is UTM, or UAV traffic management. This is one of Martin’s areas of expertise.
UTM incorporates a suite of services and safety critical functions that ensure drones can move around without inflicting harm or damage to themselves, people in the air or on the ground, and critical infrastructure.
UTM is “based primarily on the sharing of information between operators on flight intent and airspace constraints”, explained the Federal Aviation Administration Concept of Operations, and “can offer services for flight planning, communications, separation and weather, among others”.
Part of any discussion around adoption is acceptable levels of risk. How robust can a safety chain be made for a vehicle that is nowhere near the price point of a Boeing 787? And, given that it’s not carrying over 200 people, how robust does it need to be?
“We set specifications for large commercial aircraft that are extremely exacting, and if one of those hits the ground, the consequences are often severe,” Martin said.
“But a two or a five kilo UAV with no-one on board … there’s probably some latitude to commensurately scale the rigour of those standards. But by how much? That all comes back to societal acceptance of risk, because there are different thresholds for different applications.”
Risk assessment is part of Martin’s side-gig leading quantitative methods at JARUS (Joint Authorities for Rulemaking on Unmanned Systems, a global effort to harmonise technical, safety and operational requirements for remotely piloted aircraft).
“There’s a bunch of different things where it’s, ‘All right, now the safety determinations need to be risk-based rather than the historically prescriptive mechanism that has been applied in the past’,” Martin said.
“Unfortunately, that’s where it all starts to get very boring for many of the participants in the high-tech community.”
Martin added that the shortfalls in safety regulation had historically been a road block for progress, and failing to acknowledge the delays this introduces can kill a startup’s cash flow and aspirations.
JARUS is one of several professional hats Martin wears in conjunction with his role as the Applied Research lead at Nova Systems. This includes adjunct Professorships at Queensland University of Technology and University of South Australia, and being the Australian representative for NATO Applied Vehicle Technology Panel 278.
It’s a varied career that began with an RAAF apprenticeship at 16 and has included pilot training, a PhD in machine learning in speech-enabled applications (sponsored by the Chief of Army), a military fellowship at the Defence Science Training Organisation and time at NASA’s Ames Research Center as an invited researcher.
Work at Nova has earned Martin a place on the 2018 list of Australia’s Most Innovative Engineers. He is currently the Project Director for a Nova-led consortium in Singapore, overseeing a world-leading traffic management project, handling the safe delivery of people and packages and other UAV applications.
In May 2017, Nova managed a comprehensive test and evaluation for UTM and BVLOS (beyond visual line of sight) UAV flight for Queensland local government. The experience they gained during that time enabled them to successfully respond to a Singaporean government’s call for proposal, with the project subsequently commencing in October 2018.
Partners include AGI Onesky, Scout Aerial, M1 (a local telco), with support from Intel and Amazon. Importantly, the project has created a suite of jobs for both Australians and Singaporeans, with Martin highlighting that Nova and their partners have employed “some seriously bright young engineers” to support the project, such as Zi Huang and Jaya Sudarpa from Nova, and Pelwinderpal Singh at AGI.
Martin said the project intent is to extend AGI’s UTM prototype, featured at the 2018 Singapore Airshow, and expand its functionality with a mix of Nova products and know-how so that it is operationally deployable in the Singaporean urban environment.
But there are numerous challenges to overcome in operating UAVs in a metropolitan area like Singapore. One notable example is the effects of urban canyoning, where communication and navigation signals are interrupted by buildings and infrastructure.
“This environment can seriously degrade the GPS accuracy alongside the UAV command and communication (C2), in a variety of ways,” he said.
Compounding this is weather effects that can be highly variable in these canyons, particularly in tropical settings, and this can seriously degrade track keeping. Other unknowns include: “What’s your comms latency for the LTE or 4, or 5G network, in comparison to the manned environment? And under what conditions does network performance start to degrade to a point that it’s unacceptable?” And ultimately, “How far apart should you space UAVs from each other, buildings and one day, UAMs?”
Answering these questions motivates part of the modelling effort.
“We take telcos’ descriptions for describing signal properties and network latency in performance and map it to aviator speak,” Martin said.
“With that knowledge of the communication, navigation and surveillance capabilities inherent to both the UTM and the variety of vehicles that will fly within it, it’s then possible to design feasible routing that balances safety, traffic demands and efficiency expectations for supply chains.”
It is part of a multi-faceted, two-year set of trials, which could later incorporate urban air mobility (UAM) or personal air vehicles (PAVs).
“Essentially, a UTM can cater for two types of lift and shift: people and packages,” Martin said.
Exact dates are not set, but the project will conclude in mid-2020 with a multi-UAV hub and spoke delivery situation with BVLOS flights.
“Flown over people with deliveries, with Amazon injecting synthetic entities to test both the internal traffic management of our USS [UAV service supplier], but also the USS to USS interface as well,” Martin said.
“The idea is to validate the robustness of our traffic management system, particularly its ability to optimise demand and to safely deal with a variety of contingencies.”
Chicken and egg
Both the technical and the standardisation challenges are vast when it comes to the adoption of UAVs.
Martin describes it as “the industry doesn’t know what to build and the regulator doesn’t yet know what to specify … a lack of detail in the spec and an inability to know what to build.”
Martin and other professionals working in Australia have spoken highly of CASA’s collaboration with industry to address this circular problem.
For those on the industry side, he added, “My personal view is that the best way to deal with conditions of uncertainty is to experiment in a controlled manner.”
« Back to newsroom