Our friends electric

AS GIANT welding robots go about their business in a modern car factory, the scene looks like a cyberpunk vision of Dante’s “Inferno”. Amid showers of sparks, articulated mechanical arms nearly the size of telephone poles move sections of partially built vehicles so “scarily fast” that anyone who accidentally ends up in the wrong place is as good as dead, says Rodney Brooks, the boss of Rethink Robotics, a robot-maker based in Boston. For this reason, industrial robots operate in cages or behind security fences. But by segregating robots from humans, such safety measures greatly limit the tasks that robots can perform. In car factories, for example, most of the final assembly is done, expensively, by hand.

Neither workers nor robots can reach their productive potential without interacting more closely, says Volker Grünenwald, head of systems integration at Pilz, a German engineering firm. Eager to design machines that can be used for a wider range of tasks, technologists are now figuring out how to bring robots “out of the cage” so that they can work safely and more productively with people. The aim is to combine the dexterity, flexibility and problem-solving skills of humans with the strength, endurance and precision of robots. The emergence of “co-operative” or “collaborative” robots, as these new machines are called, could also lead to robots that are better able to help out in the office, at school or in the home.

Last December, in a company first, German carmaker BMW introduced a slow-moving collaborative robot in its factory in Spartanburg, South Carolina, which co-operates with a human worker to insulate and water-seal vehicle doors. The robot spreads out and glues down material that is held in place by the human worker’s more agile fingers. When this is done without the help of a robot, workers must be rotated off this uncomfortable task after just an hour or two to prevent elbow strain. Today four collaborative robots work in the facility, and more are coming, in Spartanburg and elsewhere.

BMW expects “a big, massive roll-out” of the technology in 2014 in Germany, despite the country’s tighter restrictions on human-robot interaction, says Stefan Bartscher, BMW’s head of innovation. The company plans to design additional tasks for collaborative robots as they are progressively introduced in five carmaking plants. These robots will require different technologies from those found in traditional, non-collaborative robots. Indeed, when it comes to dealing with humans, robots have so few skills that even a seemingly simple task such as handing over an object commonly ends in a tug-of-war, says Elizabeth Croft, a roboticist at the University of British Columbia.

With funding from GM, America’s biggest carmaker, Dr Croft’s Collaborative Advanced Robotics and Intelligent Systems Laboratory is designing robots that can execute “unscripted” handovers to humans. This requires the robot to determine whether a person wants and is authorised to have a particular item—be it a power drill, a document or a cup of tea. The robot must then present the item in the most advantageous orientation for the human, adjusting its grip as the object’s weight shifts. Finally, the robot must let go only when its sensors detect that the object is being purposefully and safely taken away.

Safety first

Dangerous industrial machinery is typically shut down the instant a worker “breaks” an infra-red light curtain or opens a door to enter a robot’s cage. But safety systems of this sort have drawbacks. Breaches typically stop an entire production line, alarming employees and causing delays that may cascade throughout the plant. Pilz has developed a multi-camera computer system that monitors the area surrounding robots and adjusts their behaviour accordingly.

Called SafetyEYE, the system allows a robot to, say, rivet an aircraft wing without sectioning off the entire area from people. Aware of its surroundings, the robot can roll along the length of the wing, slowing its movements if a worker approaches or, if he gets too close, stopping altogether without disrupting activity elsewhere. Since it was launched in 2007, SafetyEYE has allowed robots to be deployed in parts of factories where setting up light curtains or safety cages would be expensive or impractical.

There are additional ways to avert accidents. Some robots have red emergency-stop buttons. Researchers have even made pressure-sensing “artificial skin” by sandwiching a rubbery silicone made with carbon black, a conductive material, between electrodes. Compressing it with a slap generates an electrical signal that instructs the robot to freeze. For an additional override function, robots could be fitted with microphones and stopped with a shout, says Per Vegard Nerseth, robotics boss at ABB, a Swiss industrial giant based in Zurich which has ramped up development of collaborative robots in the past few years.

Robots capable of teaming up with people are typically used to perform tasks that are being automated for the first time, so productivity gains are especially high—provided the devices are quick and easy to program. A one-armed robot (pictured above) made by Denmark’s Universal Robots (UR) to “work right alongside employees” can be set up within an hour. Programming usually takes less than ten minutes. The user manually moves the arm and the tool it is holding from the starting point of a task to the end point, tapping a touchscreen “record” button at points along the way. Once the task is named and saved, the robot can be put to work.

Machine workshops often program collaborative robots to perform tasks for only a brief period. UR’s models can be fastened to a workbench to, say, screw together eyeglass frames to meet a rush order, and then moved to cap and box jars to cover for a worker who is off sick. Traditional robots, by contrast, are typically configured by highly paid, specialist engineers who work on a mock production line, so that the real production line need not be shut down for the weeks or months required for programming. UR sold more than 700 robots last year and expects to sell 1,500 this year, some to clients with just a few employees. Many users say that they recover the investment in a €20,000 ($27,000) UR robot within six months.

Programming collaborative robots will become even easier as software improves. Already, some experimental robots can be configured using spoken commands such as “create new skill” and “save pose”. Dr Nerseth of ABB reckons that it will eventually be possible to program robots using speech. And the control files for robots can be posted online for downloading by other users, who can tweak them as needed.

At the same time, better artificial intelligence is even rendering some programming unnecessary. Rethink Robotics says its two-armed collaborative robot, called Baxter (pictured below), uses common sense to figure out some movements on its own. Factory workers use Baxter’s touchscreen “face” to point out the objects it will handle. Baxter then studies them from all angles to determine if, say, a glass is best grasped by the outside or by inserting and opening its fingers. If a conveyor belt bringing items to be processed slows down, so does Baxter. More than 100 have been sold since the robot went on sale in late 2012.

For decades robots have been getting faster, stronger and more precise. The new breed of collaborative robots, by contrast, will move more slowly, lift less and be less precise. And yet this is the breed that will usher in the real robotics revolution, says Dr Brooks of Rethink Robotics, because such qualities will allow robots to team up with people. He points out that it was the advent not of mainframes but of less powerful but more user-friendly PCs that carried computing into the mainstream.

Collaborative robots are developing so quickly that international-standards bodies are having trouble keeping up. The world’s largest compiler of voluntary industrial standards, the International Organisation for Standardisation (ISO) in Geneva, has yet to work out safety standards for collaborative robots, such as how much force a robot can safely apply to different parts of a human worker’s body.

The ISO needs about two more years before it can publish pain-threshold standards, says Matthias Umbreit, an expert working on the project who also works as an automation specialist at Germany’s BGHM, an insurer of woodworkers and metalworkers. But the signs are encouraging, he says. A hand clamped in a robot’s gripper, for instance, can probably safely bear a pressure of 160 newtons per square centimetre. Fortunately, says Dr Umbreit, many useful tasks can be carried out using less force, so safety standards will not make robots so feeble that they are no longer useful. Johan Wahren of the Swedish Standards Institute notes that establishing standards will speed up R&D.

Friend or foe?

No matter how flexible, easy to program and safe they are, collaborative workers may not be welcomed by human workers to begin with. The experience of Alumotion, an Italian distributor of UR’s robots, is illustrative. Workers fear being replaced by robots, says co-owner Fabio Facchinetti, so his salespeople carry demonstration units in unmarked cases and initially only meet a potential client’s senior management behind closed doors.

But rather than firing workers, Alumotion’s clients often end up adding shifts because production costs drop. RSS Manufacturing in Costa Mesa, California, says its new UR robot is helping the firm compete against Asian makers of brass plumbing fixtures. Geoff Escalette, the firm’s boss, plans to buy more robots because without them some milling machines run at about 60% capacity for lack of a nearby worker able to load objects fast enough. It is worth remembering that people also used to worry that computers would steal jobs, notes Chris Melhuish of the Bristol Robotics Laboratory, a joint venture between the University of Bristol and the University of the West of England. Instead, computers helped people become more productive.

Workers generally warm to collaborative robots quickly. Employees are keen to offload the “mindless, repetitive stuff”, as one roboticist puts it. And because workers themselves do the programming, they tend to regard the robots as subordinate assistants. This is good for morale, says Esben Ostergaard, UR’s technology chief. In late 2012 Mercedes-Benz began equipping workers who assemble gearboxes at a Stuttgart plant with lightweight “third hand” robots initially designed for use in space by the German Aerospace Centre. The German carmaker’s parent company, Daimler, is expanding the initiative, which it describes as “robot farming” because workers shepherd the robots “just like a farmer tending sheep”.

Don’t frighten the humans

To keep human workers at ease, collaborative robots should also have an appropriate size and appearance. Takayuki Kanda of the ATR Intelligent Robotics and Communication Laboratories in Kyoto says that collaborative, humanoid robots should generally be no larger than a six-year-old, a size most adults reckon they could overpower if necessary. Large eyes make robots seem friendlier and, crucially, more aware of their surroundings. But overly humanoid features can lead to problematically unrealistic expectations, says Ulrich Reiser of Fraunhofer IPA, a manufacturing research institute in Stuttgart that makes a €250,000 home-assistant robot called Care-O-bot. He notes that people tend to distrust robots with protruding sensors, “Terminator”-like exposed cables, or a jerry-rigged, student-project look.

To interact smoothly with people, robots will also need “social intelligence”. It turns out, for example, that people are more trusting of robots that use metaphors rather than abstract language, says Bilge Mutlu, the head of the robotics laboratory at the University of Wisconsin-Madison. He has found that robots are more persuasive when they refer to the opinions of humans and limit pauses to about a third of a second to avoid appearing confused. Robots’ gazes must also be carefully programmed lest a stare make someone uncomfortable. Timing eye contact for “intimacy regulation” is tricky, Dr Mutlu says, in part because gazes are also used in dialogue to seize and yield the floor.

When a person enters a room, robots inside should pause for a moment and acknowledge the newcomer, a sign of deference that puts people at ease, says the University of British Columbia’s Dr Croft. Robots also appear friendlier when their gaze follows a person’s moving hands, says Maya Cakmak of Willow Garage, the California-based maker of the PR2, a $400,000 robot skilled enough to make an omelette—albeit slowly.

It will probably be a decade or two at least before the descendants of PR2, Care-O-bot, and other “home assistance” or “companion” robots will be nimble and intelligent enough to zip autonomously through houses performing chores. They will need far better sensors, movement-control actuators and batteries, and much smarter software. They must also be capable of displaying empathy or they will be rejected, says Kerstin Dautenhahn, head of a “social robotics” team at the University of Hertfordshire in Britain.

Her team’s Care-O-bot robots crunch data from 60-odd household sensors that monitor door and cupboard hinges, taps, electrical appliances and so forth. If medicine isn’t taken, say, the robot may alert relatives or the hospital. It is vital that a robot of this sort is not perceived as hostile, but as having its owner’s best interests at heart.

“To keep humans at ease, collaborative robots should have an appropriate size and appearance.”

One way to do this is to give robots a defining human trait—the ability to make mistakes. Maha Salem, a researcher under Dr Dautenhahn, programmed a humanoid Asimo robot, made by Honda, to make occasional harmless mistakes such as pointing to one drawer while talking about another. When it comes to household robots, test subjects prefer those that err over infallible ones, Dr Salem says.

Another approach uses sensors to assess the state of nearby humans, so that robots can respond appropriately. With funding from the European Union, researchers are using bracelets equipped with electrodes to enable classroom robots to determine whether students are bored, confused or anxious. The robots can adapt their teaching style accordingly, says Iolanda Leite of the Instituto Superior Técnico, a Portuguese university participating in the programme, which is called EMOTE. One of its objectives is to foster “social bonding” between people and robots.

Such bonding could have some surprising uses. In experiments carried out at Yale University involving a biped humanoid called NAO, made by a French firm called Aldebaran Robotics, children proved to be just as willing to share secrets with it as they were with an adult. The researcher who performed the experiments, Cindy Bethel, who is now at Mississippi State University in Starkville, has also found that children who have witnessed a crime are less likely to be misled in a forensic interview with a robot than with a human expert—even one trained to obtain testimony. Mark Ballard of the Starkville police department, who has been working with Dr Bethel, reckons that the robots needed to conduct “child friendly” forensic interviews will be available by 2020.

What’s next? Market research is not much good at predicting developments in the field of collaborative robots, says Bruno Bonnell of Robolution Capital, a robotics investment fund in France. For one thing, he says, people say they want complete control over robots, but once they start using them they actually prefer them to be as autonomous as possible. Working alongside robots changes the way people think about them, in other words. Whether on the factory floor, at home or in the classroom, the evolving relationship between human robots will be defined by a process of collaboration.