Marc Hanheide and Linda. Images by the author
If you imagine a robot security guard—an electronic watchman that can patrol corridors, keeping an eye out for anything suspicious—you probably wouldn’t picture Linda. About the height of a small adult, it looks a bit like an elongated Dalek but is conspicuously lacking any weaponised appendages. In fact, it has no arms; just a touch screen, a whole host of sensors, and a pair of long-lashed but unseeing googly eyes.
I “met” Linda the robot at the National History Museum in London, where it’s showing off its security patrolling skills as part of Universities Week. The robot moves autonomously around one of the museum’s hallways, navigating the changing environment as it goes and completing various tasks. It can roll around dodging the hordes of visitors on its patrol, detect “intruders” in one of the quieter rooms, and, rather more frivolously just for the sake of the exhibition, take your picture and tweet it.
It’s all very cute watching the robot try to wiggle its wheels through a crowd of overexcited schoolchildren, first attempting to slowly nudge a path and then politely requesting that people move, but it’s also an impressive illustration of how robots like Linda could be employed in the real world. Some of its comrades—there are six in total, alike except for colour and name—have already been deployed in trials at a care home in Austria, and at the UK headquarters of G4S, the private (and controversy-beleaguered) security firm. This bot has a serious side.
Linda 'patrolling' a corridor at the Natural History Museum in London
Linda is part of a European project called STRANDS, or Spatio-Temporal Representation and Actvities for Cognitive Control in Long-Term Scenarios, which set out last year with the objective of making robots that can run autonomously for a long period of time in regular (read: changeable) human environments. Marc Hanheide, a senior lecturer at Lincoln University and one of the principal investigators on the four-year project, told me that he and his team were frustrated at seeing robots “designed to work for just 20 minutes to show off their research thing,” without having a chance to really learn anything.
On the other hand, many industry robots run 24/7, but they only do the same simple tasks. “You think, they do this for a living, that’s what they do all the time—they should learn about this at some point,” he said. “But they don’t; they do the same stupid thing over and over again.”
The STRANDS robots are designed to last for at least a few weeks (one year into the project, Linda can function autonomously for 15 days) and actually make use of the experience they gather over that time. By observing the world around them, they can adapt to changing environments and, most importantly, recognise any deviation from routines—like an intruder snooping around a restricted area, or an elderly resident fallen in a corridor.
“The only thing it can basically do is move from one place to another, but it does that quite smartly,” Hanheide explained. Linda continuously makes 2D and 3D maps of its surroundings using a laser sensor where you might consider its ankles to be, a Kinect-like sensor on its head, and a second on its body pointing downwards to detect potential disturbances like cliff edges. Using these sensors, it moves autonomously (no remote control needed) and observes changes: a person walking by, a door opening or closing, an object that wasn’t there before.
A representation of one of the maps Linda makes and the route it has calculated (green). The red dots show obstructions to its laser sensor (like people's legs)
Thanks to machine learning algorithms and predictive probabilistic models, it can use a kind of “common sense” to work out what it’s likely to be observing, and how unusual the activity is for that time and place. It can then send a message if something’s amiss. All of the data it collects stays onboard in three PCs.
To give an example of a specific task, one of the things the robot was asked to do while at G4S was to check workers’ desktops at the end of the day. Hanheide explained that company policy required desks to be clear of paper, so Linda learned what the clear desks looked like and could then flag anything that seemed out of place.
Intruder detected! (Actually Jaime Pulido Fentanes, a research assistant at the University of Lincoln)
Where Linda differs from less “sensible” robots is it’s not limited to specific tasks. If you command it to check if certain doors are closed, for instance, it’ll do it—but it’ll also let you know if the bin next to the door is on fire, or anything else out of the ordinary. You know, like a human would.
But Hanheide said he doesn’t see robots like this stealing security guards’ or carers’ jobs—after all, it can only flag problems, which a human needs to deal with. He can, however, see them working alongside regular staff. “For some of the very mundane checking tasks that guards do, this could be an alternative,” he said. “It could also be an alternative to covering a whole building with surveillance cameras.” The group has already had a call from a “big company” Hanheide was unable to name about using a robot like Linda to patrol their headquarters.
Robots, of course, have a couple of advantages over humans. For one, they don’t get tired or bored. To keep Linda and its pals autonomous for long periods of time, the team made it so they could manoeuvre themselves onto a charging station to recharge themselves, and also programmed them to ask for help when something inevitably goes wrong. Hanheide recalled one expedition when the robot got stuck in a thick carpet. It first looked around for someone and called out for help, then sent a text to ask for assistance.
In other words, they're not going to be working completely unsupervised anytime soon; they're built for collaboration with humans. You can interact with Linda via its touchscreen, and it also has a microphone and speaker (though the team isn’t focusing on voice recognition). As a kind of symbol of this human-robot cooperation, its eyes don’t actually do anything in terms of vision, despite the fact that they blink. They just move to show the robot’s intention, such as which direction it’s about to head in—a kind of non-verbal signal to help us humans figure out what our soon-to-be robot colleagues are up to. If we're going to be working alongside them, we'll have to at least know how to communicate with them.