Four hours after American Airlines Flight 11 hit the North Tower of the World Trade Centre on September 11, 2001, iRobot - a small robotics firm based in Bedford, Massachusetts - received a call from the US military's secretive advanced technology program, known as DARPA (Defense Advanced Research Projects Agency). Staff at iRobot immediately broke into teams to make last-minute modifications to four robots, packed a van and headed south. By midnight they were at Ground Zero in Manhattan, known then only as "the pile".
Over the coming days and nights, iRobot's machines used their infrared sensors to search for survivors in buildings that were still standing but too dangerous for rescuers to enter. No one was found, but the robots had proved what they could do.
A month later, the US invaded Afghanistan and troops of the 82nd and 101st Airborne Divisions were equipped with the same robots. Rodney Brooks, the Australian former Massachusetts Institute of Technology professor who founded iRobot and led the development of the so-called PackBots, says the soldiers were initially sceptical. "[They] said, 'We don't need those stinking robots', until they saw those black holes at the caves and said, 'How about we send a robot in?' " Before long, whenever an airborne chopper left Bagram Airbase on operations, a seat was reserved for a robot, says Brooks.
Since then, PackBots have worked alongside soldiers on battlefields across Afghanistan and Iraq. Over the years, and as the technology behind the robots has improved, their handlers have found them new tasks. Following Japan's Fukushima nuclear power plant meltdown in 2011, iRobot's phone rang again. This time it took them 48 hours to deploy their robots into a facility so old that emergency workers outside had no idea what the analogue dials on control panels inside the plant were revealing. PackBots entered the plant, set up a Wi-Fi relay and put cameras onto the control panel. They are still in there today, says Brooks.
Brooks has since left iRobot and set up a new company, Rethink Robotics, which is developing just one robot called Baxter. Baxter has never been to war, but is attracting far more attention - and in some quarters consternation - than any other machine Brooks has built.
Rodney Brooks was born in Adelaide in 1954, four years after US author Isaac Asimov's I, Robot was published. The book popularised the notion of autonomous humanoid machines living and working among us. Brooks' father, Harry, worked on rockets at the Woomera Test Range, then an Anglo-Australian facility. Brooks was always fascinated by technology, an interest fostered not only by his father but also by the How and Why Wonder Books series that were published in the 1960s and '70s.
With instructions from the books, he was soon making basic machines out of wires, nails and light bulbs. At 12 he built one that could win at noughts and crosses. A few years later he saw Stanley Kubrick and Arthur C. Clarke's 2001: A Space Odyssey, and the possibilities of robotics dazzled him.
By this time, the first industrial robot was working in a General Motors factory in New Jersey, moving hot metal from a die-casting machine. But in pop culture, robots were still portrayed as fantastical talking machines like the one that flailed its arms about in the '60s TV series Lost In Space. By contrast HAL, the robot in 2001, made so much more sense to Brooks. It was intelligent, responded to its environment, understood voice commands and could speak with its operators. It demonstrated the possibilities of what became known as artificial intelligence. Brooks was captivated and still refers to the film in public lectures.
"It [HAL] did happen to be a murdering psychopath, but apart from that it was great," he says.
Brooks graduated dux of Plympton High School - "I was a pain in the arse" - and went on to study pure mathematics at Flinders University, where computer science did not yet exist. While there, Brooks was able to get his hands on a computer for the first time, a card-driven IBM 1130 with 16 kilobytes of memory that had four full-time operators during the week. At the end of his first year, Brooks and another student were given the machine to play with for 12 hours every Sunday: "We learnt everything about it. We built AI [artificial intelligence] systems on punch cards."
Brooks went on to get his PhD in computer science at Stanford University, a hub for the technology industry, and later held research and faculty positions in the heart of America's robotics world at Carnegie Mellon University and the Massachusetts Institute of Tech-nology (MIT). Today he is one of the world's leading roboticists, a handful of people responsible for taking the field from the technological backwaters to the forefront of a new industry. He speaks softly, but clearly and passionately. When he's on stage demonstrating his machines, he has that easy energy that marks him not just as a teacher but as an evangelist.
However, 30 years ago, just as Brooks started pushing the boundaries of robotics, it seemed everyone else had lost interest. In the 1980s, robots were starting to look a bit old-fashioned. In research universities, the internet had taken off and cyberspace was being built and charted. While a generation of talent was drawn to Stanford and Silicon Valley to work in the bright Californian sun, Brooks and a handful of die-hards continued their lonely research in places like MIT just outside Boston, on the cold, grey east coast.
It was here amid this isolation that Brooks made a breakthrough that would change robot development around the world. It had always bothered him that no matter how much computers improved and the brain capacity of robots increased, they remained fairly stupid and clumsy machines. "I had been fascinated by how insects with tiny little nervous systems could do way better than any of our robots could, yet we had the biggest mainframes," he says.
The insight evolved into a paper published in 1985 called "A Robust Layered Control System for a Mobile Robot." Brooks remembers overhearing a senior scientist saying, "Why is this young man ruining his career?" before he presented the paper at a conference in France. Its key insight was that robots were slow because all the information about the environment that they absorbed from sensors - such as, say, cameras - was being channelled back into their computer brains for real-time processing before the machines were able to react. Even as computers improved, the process remained inefficient.
Brooks proposed separating higher- and lower-order thought in artificial intelligence. At a base level, a robot moving about a room might be following an instruction to avoid an object. At a higher level, it might be following an injunction to explore its environment. This would, crudely, replicate a simple nervous system.
Brooks believes this basic separation of tasks mimics the evolution of the brain. His insight grew into the field now known as behavioural robotics, and he and his team at MIT went on to build ever more sophisticated robots with the support of NASA and DARPA.
By now he was sure his belief that one day we would be living and working alongside robots was right. Alongside his duties at MIT, Brooks founded iRobot with a couple of his students to help make that happen - which brings us to the revolution that occurred when nobody was watching.
Operation Enduring Freedom was launched in Afghanistan on October 7, 2001. Soon after, the Taliban collapsed and the focus of the fighting turned to the fortified caves in the country's Tora Bora mountains. PackBots were sent in, and what they achieved was extraordinary for autonomous machines. They proved they could not only function in the tunnels and the dusty roads, but that they could search for and find booby traps and detect explosives. More importantly, they proved they could do so in combat conditions under the direction of young soldiers with no specialised training - often with no education beyond high school - who themselves were under fire. Over the following years they were deployed across Afghanistan and Iraq, and today iRobot still receives thank-you cards from the soldiers who worked with them.
Their success comes back to the massive advances in behavioural robotics. The soldiers did not have to program the robots to make the decisions needed for a machine to pick its way along a dark tunnel; they merely ordered them to do so. Using their various sensors, the robots worked out how to do it.
This was a crucial turning point, says Brooks, but not the only one that happened at that time. Even as it was developing the PackBot, iRobot was working on a simpler machine, the Roomba, a small circular domestic vacuum cleaner that was first released in 2002. Using just two sensors, a Roomba can find its way around its workspace - a floor - and back to its charging station. It can identify dirty patches that need extra cleaning, and map and memorise where it has been and where it needs to go. It can also avoid objects and stop itself from tumbling down stairs.
The Roomba was an ideal product for iRobot to explore the possibilities for domestic robotics - vacuuming is a simple and necessary task - and introduce robots into homes. Even its price was significant. Other vacuuming robots existed when Roomba was developed, but they cost thousands compared to Roomba's hundreds. If robots are to be of use, their cost has to be appropriate to their purpose. Today, 10 million vacuum-cleaning robots have been sold worldwide, with the Roomba leading the charge, and each generation is smarter than the last.
Taken together, Brooks believes, the introduction of the PackBots and the Roomba constituted a turning point in the evolution of robots, albeit one that few roboticists noticed at the time.
By 2008, however, Brooks was ready for another challenge. He still wanted to create a robot that worked alongside people in an industrial setting - one that might evolve into the type of household robot that featured in the books and cartoons he grew up reading. He left iRobot and took leave from MIT to focus on his new project - the robot that is causing so many people such consternation, the one he has named Baxter.
When I first meet Brooks at Rethink Robotics headquarters, Baxter is standing at a workbench. He is about my height, 178 centimetres, and has kind, expressive eyes on a flat-screen face. We play a game of Connect Four together. I go first and Baxter looks intently at the move for a few seconds before he slowly reaches for his own tile from the pile between us. His claw hand manipulates the tile with surprising delicacy.
Brooks has banned his staff calling Baxter "he" or "him", although they ignore the rule. Brooks' concern is not one of scientific pedantry; he just doesn't want the product imbued with a gender.
"You shouldn't have called him Baxter then," I say. Brooks explains that "Baxter" emerged from the non- gender-specific "baker". He seems to spend a lot of time doing this: explaining in logical terms things that are obvious to him, but aren't to anyone else.
"Of course I call him 'he', " an engineer tells me. "We spend all day with him, he is our workmate. He is Baxter." Unlike most other robots on the market, Baxter is designed to work alongside people in small- and medium-sized businesses, doing the same sort of work that people do. He - it - is designed to be interchangeable with an unskilled worker.
With their fast-moving steel arms, robots in heavy manufacturing industries are penned-off behind safety barriers. Baxter's two arms move slowly and are sheathed in soft pliable plastic. A manufacturing robot might be programmed to drive a rivet through a plate. To achieve the task, its arm is directed via co-ordinates to a fixed point where it is then directed to perform an action. Baxter is designed to be directed to do a task, then work out how to do it himself.
To this end, he has proven most useful for tasks that involve packing. To direct Baxter to pick up a box from one conveyor belt and place it into a bag on another, an operator simply needs to move his own arm through the series of motions while holding a button on Baxter's cuff.
Teaching Baxter to do a new task only takes a few minutes and can be done by a factory worker with basic training. What's more, because Baxter knows to an extent what he is doing, he can correct errors. Brooks shows me what happens when he pulls a parcel from Baxter's hand. The robot pauses, realises something is wrong, and goes back for another package. Other robots would simply continue with the task.
Baxter can use both arms simultaneously to undertake separate tasks, or use them in tandem for more complicated operations. He can, for example, pick up an item, weigh it, discard those that don't meet a specification and pack those that do. Baxter's face and those kind eyes are not just for decoration, either. Baxter turns and looks at his tasks, alerting people working around him to what he is doing.
His face has several expressions, which also serve as signals to human workmates. Should a worker get in Baxter's way, contact sensors in his arms stop them from moving. Baxter can't accidentally hurt you. He is equipped with cameras in both wrists and in his head. He also has an infrared sensor to give him spatial awareness. When you approach him, a ring of LEDs lights up above his face, letting you know he is awake and aware.
But perhaps Baxter's most significant feature is how easily he can be taught a new task. Most robots in factories need a trained operator to program them to perform new actions.
Rethink Robotics' head of marketing Sue Sokoloski tells me about a factory that has bought a Baxter unit. It works alongside a woman who has no advanced training and speaks English as a second language. But she is responsible for teaching the robot new tasks.
Freed from the packing conveyor belt, the worker is able to do other more challenging - and interesting - work, says Brooks. Far from being a threat to the jobs of manual workers, Baxter will relieve them of work that is "dull, dirty or dangerous", he says.
One of Rethink's early investors was Jeff Bezos, the founder and chief executive of Amazon. Brooks stresses that Bezos's investment was personal rather than a company one, but Bezos's interest in new technologies is long-standing. Earlier this year he announced he would like to use drones for deliveries.
It is not hard to see how a robot like Baxter might one day fit into Amazon's business plan. Amazon employs 40,000 warehouse workers - out of a total of 117,300 full- and part-time employees - who are paid an average of $US12 per hour, or about $US25,000 per year. Rethink is reluctant to talk about the cost of Baxter, but it has been reported to be between $US22,000 and $US25,000. For that, an employer gets a robot that can work around the clock without making the mistakes common to workers in repetitive jobs, such as counting and measuring. Nor do robots get bored or sick, join trade unions or, as one analyst dryly noted, take legal action against their employers.
Amazon has spent more than $US14 billion since 2010 building new warehouses around the United States, intended to enable the company to deliver most items the same day they are ordered. Storage, shipping and delivery of items are now Amazon's top operating expense and the company is hiring thousands of new workers to staff this vast delivery system. Despite Amazon's expansion, earlier this year Bezos announced a "Pay to Quit" program, which gives warehouse employees up to $US5000 if they agree to quit their job.
Brooks says genuinely disruptive new technologies tend to increase productivity so much that they create wealth and jobs rather than destroy them. I put this to Jerry Kaplan, a scientist, futurist and entrepreneur who teaches at Stanford University and he laughs long and loud. "That is just bull," he says.
Kaplan believes Baxter is at the forefront of a new robot workforce that has the potential to annihilate whole classes of human labour. He uses the example of American farm labour, which 200 years ago constituted more than 90 per cent of the US workforce but, with the advent of automation, has shrunk to about 2 per cent. Because the change happened gradually, the economy adjusted to absorb those workers into new fields. But with Baxter's capacity to sense and interpret the world around him and with the speed at which cheap computing capacity is increasing, that same revolution could happen within the next 10 to 20 years.
Kaplan believes robots like Baxter could soon be doing any basic physical human labour, from cutting hair to digging ditches to painting houses.
"Anything you can think of, things you never even thought of," he says. "Karl Marx was right. In the battle between capital and labour, capital will win."
In this case, that means those with the capacity to buy, build and deploy what he calls "forged" labour will reap the benefits at the cost of workers. Kaplan is no Luddite, though. He just believes we should embrace the technology with our eyes open to the economic and social risks it presents.
Brooks understands such concerns, but says people tend to overestimate the short-term impact of new technologies, even while they underestimate the long-term ones. In one of his TED talks, he uses the 1957 Spencer Tracy movie Desk Set to explain. In the film, Tracy plays a consultant installing a computer in the library of a publishing company. The librarians fear the worst, but soon discover the machine is useless without them and they are faster with it.
Brooks says the number of librarian jobs in the US increased from 60,000 in 1957 to 213,000 in 2009, when it started to decline, not because of mainframe computers, but because of the internet being carried around in handheld devices. Brooks says the short-term impact of robots like Baxter could be to cut costs enough to spur further "on-shoring" of manufacturing jobs, a phenomenon occurring as the gap between Chinese and US wages shrinks.
Brooks goes as far as saying his fear is not that there will be too many robots in our near future, but too few - too few to replace ageing manufacturing workforces, and too few to help us care for ourselves in old age, drive us to the shops, carry our groceries, cook our meals and do our dishes. For a robot to achieve this, says Brooks, they will need to develop four basic but complex capabilities: the object-recognition capacity of a two-year-old child; the language capability of a four-year-old; the manual dexterity of a six-year-old; and the social understanding of someone who recognises that some world views may differ to their own.
Brooks thinks it's possible that Baxter may evolve into such a machine - since the first unit shipped in early 2013, software updates have made his movements twice as accurate and nearly twice as fast - but he also recognises that it is difficult to predict how technology will evolve. He was one of the few who, 30 years ago, believed one day that everyone would use computers. He did not predict that everyone would eventually carry them in their pockets.
I marvel at Baxter while we play Connect Four. He identifies the tiles, picks them up, then slots them home, his big, clumsy paw finding the perfect centre of the small slots before he drops them in. I am disconcerted that he is actually trying to win, even if he doesn't know why. It takes longer than I expect to beat him and, when I do, he stares at the plastic grid for a long moment before shaking his head from side to side, his eyes curled into a sad frown.
Lead-in photograph by David Yellen.