16 September 2022
Amalyah Hart
Amalyah Hart has a BA (Hons) in Archaeology and Anthropology from the University of Oxford and an MA in Journalism from the University of Melbourne.
Deep in the bowels of the Earth, some thirty metres below the bustling city of Louisville, Kentucky and unbeknownst to most of its 600,000 residents, an epic contest is about to unfold.
Gathered on the cool stone floor of the massive Louisville mega-cavern, a sprawling former limestone mine that now rates as Trip Advisors’ #1 local attraction, a team of CSIRO-designed robots are waiting for their cue.
Some of them are hardy-looking, all-terrain bots with shiny red casings and caterpillar-like tracks that can wheel them along over difficult ground. Then there’s a pack of vivid-yellow quadrupeds, like a gang of freakish future-dogs, loitering alongside.
A nervous team of CSIRO roboticists, who have been preparing for this moment for three long years, wait their turn. On the signal, it’s time to send the bots in.
The DARPA Subterranean Challenge (known as the DARPA SubT) is often described as the ‘robot Olympics’, and Olympian it is. Set up by the Defence Advanced Research Projects Agency, a wing of the US Department of Defence, the challenge involves a series of complicated tasks, set over three years and three different terrain types, in which cutting-edge robots must rapidly map, navigate and search sprawling underground networks for hazards, lost or injured humans, and items of value.
The contest, which had its Louisville final across four days in September 2021, is designed to mimic real-world settings, like military zones or natural disasters – settings that are often extremely dangerous for human personnel, and which the rapidly growing field of robotics could revolutionise.
The “systems competition” – in which the CSIRO team, known as Data61, competed – involves the navigation of three subdomains: a tunnel system, an urban underground network, and a cave system.
“The task was essentially set up as a scavenger hunt,” says Dr Navinda Kottege, leader of the Data61 team and Group Leader for Robotics and Autonomous Systems at CSIRO. “Robots go into the course, spend an hour looking for a series of predefined objects, like thermal mannequins representing human survivors, and you need to correctly identify the object and its location.”
Those hard-working robots, worth tens of thousands of dollars apiece, are rewarded for their service with a lengthy and cold entombment.
Scavenger hunt is perhaps an anticlimactic description of the challenges robots in the DARPA SubT face. They must find their way through rough terrain, fending off dangerous obstacles like pockets of carbon dioxide gas. Sometimes, the entry-way will close off behind them like in an Indiana Jones film, or a passage will suddenly fill up with smoke. Sometimes, impenetrable rock scuppers communication, leaving the machines to fend for themselves.
That’s where Data61’s expertise comes in: as Kottege tells me, the robots are bought commercially from major producers, but the programming that makes them smart, and able to make their way autonomously, is done by the competitors.
Once the objects have all been correctly identified, the competing teams don’t tend to bother prizing them back out of the course: it’s all about the data they’ve gathered. Those hard-working robots, worth tens of thousands of dollars apiece, are rewarded for their service with a lengthy and cold entombment.
Data61 won big in the 2021 DARPA SubT, pulling up second on the podium, and you can watch the full run of their final event online. Now, CSIRO has announced that the prize money is now being reinvested in a $1.2 million scholarship fund at the Queensland University of Technology for future industry leaders in robotics from under-represented backgrounds.
What is the future of robotics, in Australia and beyond?
To understand the future of robotics, you’ve first got to understand what a robot actually is – says Professor Elanor Huntington, Executive Director of Digital, National Facilities & Collections at CSIRO.
“Everyone has a mental model for what a robot is,” Huntington says. “The way that roboticists think about robots is that they are these machines that have the ability to sense their environment, move through space, and manipulate physical objects.”
But it’s not so simple: a robot isn’t always a mere machine – sometimes, into those hard metal casings, we can actually program a sense of agency or reason.
“There’s a general concept of something called ‘agency’ which is about sensing the world, reasoning on that, and then making changes in the world according to your reasoning,” she says.
This is actually a fierce debate in the realm of scientific ethics and, if we’re getting really philosophical, the jury’s out on whether robots can possess agency (or indeed what agency really means), because of the thorny old problem of ‘free will’. But that depends how you define agency and, in software engineering, an ‘agent’ does not necessarily have to possess rights or responsibilities.
Crucially, says Huntington, “it gets to the difference between agency and autonomous or semi-autonomous.”
A robot isn’t always a mere machine – sometimes, into those hard metal casings, we can actually program a sense of agency or reason.
I ask Huntington the difference: “Well, autonomy means once you’ve told them what to do, its job is just to do it. It’s not necessarily making decisions about, ‘do I think that a production line is a good idea?’ So, there’s less of that sense of reasoning and making decisions about whether or not it’s the right move.”
All this is more than a philosophical exercise: it helps us understand what robotics can actually do for the human world.
That’s because that core difference, between an autonomous robot on a production line and a sensing, reasoning robot in the sorts of uncontrolled environments replicated by the DARPA SubT, get to the heart of what new robots could do for us.
“Those uncontrolled environments are where Field Robotics comes in,” says Huntington.
What is field robotics, and why is it so important?
Field Robotics, of which the DARPA SubT event is the ultimate pinnacle, is arguably the most exciting branch of robotics. Field robots must be robust, adaptable, and clever. They have to navigate scenarios that humans often find complicated – disaster zones, military exercises, underground mine shafts or raging bushfires. They could also seriously reduce human mortality in these high-risk contexts.
Australia, a country with vast tracts of agricultural land, escalating natural hazards and a booming resources industry, is understandably a world-leader in field robotics – and uniquely poised to benefit from this robot-boom.
“Australia’s large land mass and sparse population, as well as the prominence of certain industries such as mining and agriculture, has shaped this direction,” explains Professor Ian Manchester, a professor of mechatronic engineering at the University of Sydney and co-director of the Sydney Institute for Robotics and Intelligent systems.
Field robots must be robust, adaptable, and clever.
“There is a huge potential for robotics to assist in response to bushfires,” says Manchester. “This ranges from surveillance for spot fires and monitoring of known fire-fronts, all the way to robotic waterbombing systems.
“Firefighting generally, and aerial waterbombing specifically, are obviously extremely dangerous activities and if unmanned systems can do the job that is enormously beneficial.”
Another of Australia’s major challenges is growing enough food in a world that is warming, on a continent increasingly besieged by weather extremes, and in a time of deepening food insecurity.
“Agricultural robotics is rapidly growing in terms of demonstrated capabilities as well as start-up companies, and Australia is a world leader in this area,” Manchester says. “Robotic systems can continually monitor crops, collecting data such as high-resolution imagery and 3D scans, and performing on-board analysis of soil samples.
Australia is understandably a world-leader in field robotics – and uniquely poised to benefit from this robot-boom.
“Robots can also perform high-precision spraying, mechanical weeding, and other technologies that minimise the need for wide area spraying with fertiliser or pesticides, reducing environmental impacts such as run-off into waterways.”
Robots are also already a prominent feature of Australia’s controversial mining sector.
“Australia puts lots of robots out in the mines,” says Huntington, “we’re world-leading there. And in fact, in terms of the intersection of robotics and safety, the Australian mining industry is a thought leader around the safe use of autonomous systems.”
And field robots can be used for all sorts of other ends: in Tasmania, an exploratory robot affectionately named Wombot was sent down a series of wombat burrows to gather data on a devastating outbreak of mange caused by the long mite, Sarcoptes scabiei.
Robots are also already a prominent feature of Australia’s controversial mining sector.
But will all this automation make human jobs obsolete?
“There have been a lot of reviews that spoke to the fact that there is a transformation and a disruption, not just coming but already here, in a large number of industries and job sectors,” says Huntington.
“Those numbers can be understood as saying, ‘the robots are coming for my job’, but for the most part what people are really suggesting is human-robot teaming, where the dirty, dangerous, dull, dark work that people do will get absorbed by robots, freeing us up to do other things.”
Why is diversity in robotics lacking – and why is it so important?
Diversity is the thorn in the technology sector’s side: “robotics, like many STEM areas, has not had a very strong track record when it comes to gender diversity and other dimensions,” admits Manchester.
It’s estimated women make up between 7% and 19% of robotics engineers are female, and around 67% of robotics engineers are white.
It’s not just about sourcing talent in the first place: in fact, according to Huntington, where much diversity is lost in robotics is through attrition rates, because people from under-represented backgrounds can find it a hostile world in which to work.
Diversity is the thorn in the technology sector’s side.
“One of the things about being in the minority is it can be a lonely experience,” she explains. “And I speak from personal experience there.”
This is a problem for a variety of reasons, the most mercenary being a major skills shortage.
“There is known to be a massive skills shortage out there at the moment,” says Huntington. “There’s a global statistic that suggests that there might be 133 million jobs created around the world in the next five years as a result of the kind of new robotics technology, so where are those people going to come from?”
“So just in sheer pragmatic terms we need we need greater diversity. If all you ever do is pour on the 48-ish percent of people that we have traditionally thrown on then that’s a problem, so we can immediately double the number of people who might go into the jobs by being more gender diverse, for example.”
But that’s not the most compelling reason to expedite the shift towards a more diverse robotics workforce.
“There’s lots of evidence these days that says that diverse teams make better decisions,” Huntington explains. “And there’s lots of research that says that you get more creative teams, who find different problems to solve.”
Data61 benefits from this, says Kottege: “We have very diverse and vibrant backgrounds in our group… we have people from all parts of the world coming in and that definitely has an effect on the level of creativity and the type of decision-making.”
Beyond pragmatism, there’s a strong moral imperative to make robotics more diverse, especially as its reach extends further into our lives.
“If robots are going to be threaded through our daily lives in really mundane, practical ways, then I would like to be sure that everybody has the opportunity to be part of making the world that we’re all going to live in,” says Huntington.
That’s important, because a lack of diversity in tech can have all manner of exclusionary consequences.
In a now-viral video filmed at a Marriott hotel in Atlanta, Georgia, an African-American guest, TJ Fitzpatrick, found that he couldn’t activate any of the remote-controlled soap dispensers in the bathroom. When his white friend, Larry, put out his hand, the soap came out immediately.
Huntington says this particular flaw has been programmed into all sorts of sensing technologies, because the machines don’t account for the difference in light levels – a major programming oversight.
Another example Huntington raises is the cameras in various smartphones.
“There are a lot of algorithms that sit between your camera and what gets recorded as the photograph,” she says. “Certain phones make a more realistic rendering of what people of colour look like than others, and it’s because of the algorithms that sit inside the computer.
“So, to just live a life where you are not represented accurately in the photographs that you take on your own phone, that just feels exclusionary, right?”
And software racism goes beyond the level of micro-aggression: in May 2016, a shocking (or perhaps not-so-shocking) report by ProPublica found that a computer program used by a US court to assess the risk of a defendant reoffending was biased against black prisoners. The program was far more prone to flag black defendants as likely reoffenders, wrongly flagging them at almost twice the rate of white people.
Beyond pragmatism, there’s a strong moral imperative to make robotics more diverse, especially as its reach extends further into our lives.
In fact, a study published in June this year deliberately sought to find whether AI would be prone to racist and sexist stereotypes: the results were damning, with the algorithm displaying clear biases when sorting individual faces into categories such as ‘criminal’, ‘homemaker’ and ‘janitor’. The authors point out that new algorithms are often built on top of older ones, and any forthcoming AI needs to be screened for these toxic biases – and reprogrammed if necessary.
“The range of problems we look to solve with robots is limited by the people who identify with those problems,” says Dr Sue Keay, who runs the Australian Centre for Robotic Vision at the Queensland University of Technology and chairs the Robotics Australia Group.
“So, if men are exclusively the developers of robots, will those robots solve problems relevant to women, or will we see a repeat of the mistakes we have seen with other technologies, like the Apple health app, which ignored women’s health issues like period tracking.
New algorithms are often built on top of older ones, and any forthcoming AI needs to be screened for these toxic biases.
“I wonder what technologies we have missed out on because there have not been a diverse range of people developing them.”
Huntington says there’s lots that can be gained in the programming of robots and artificial intelligence if people from diverse backgrounds have an equal seat at the table.
“A really interesting example is that because of the dominant religious background in Japan, Shintoism – which has always had a sense of animism – the way that Japanese culture interacts with robots is really very different to the way that Western culture interacts and imagines what robots might be able to do.”
That’s the ethos behind CSIRO’s new scholarship, the Alberto Elfes Memorial Fund.
“It’s not a silver bullet solution for this problem,” says Kottege, “but we have to try and do our best to move the needle a bit in that direction.”