When you visit Manuela Veloso at Carnegie Mellon University, you’re not guided to her office by a security officer or even issued instructions by a secretary at reception. Instead, one of Veloso’s autonomous CoBots (short for “collaborative robot”) guides you to her desk — or wherever in the building she may be. It sounds like something out of the movies, and it may feel like science fiction, too, when you’re being chaperoned by a robot wheeling around on its own volition.
The CoBots, which run a combination of C++, Python, and Java, aren’t much to look at. Each of Veloso’s robot minions is a less-than-elegant assemblage of a camera and a laptop atop a wheeled base, along with a Microsoft Kinect that’s used for navigation and obstacle avoidance (see video below). A tangle of visible wires connects everything together. But function is the focus here. You assign tasks to the CoBot via a web interface, and once a task is completed, you press a button to release the robot from the task so it can either return to its home base, or embark on its next journey.
If Veloso’s CoBots tell us anything, it’s that Hollywood-style science fiction portrays robotics in a way that’s completely dissonant with where technology stands today. We’re a long way off from seeing C-3PO and Wall-E criss-cross the universe (or heck, even a city or an office space) without getting tripped up on physical obstacles. Our world isn’t static. Human interactions — and the spaces we reside in — are constantly changing and evolving.
It’s difficult for robots to anticipate and handle these physical unknowns, but Veloso, a computer scienceprofessor at Carnegie Mellon, aims to change that through her work with autonomous robots designed to perform daily, human tasks. Obstacle avoidance may not seem very glamorous at first glance — most people want their robots to shoot lasers or cry “danger!” — but it’s a linchpin of real-world robotics, and it’s one of Veloso’s key areas of focus.
Indeed, autonomous robots have significant limitations when it comes to successfully getting around on their own. Due to physical limitations (how well would you fare on rollers?) and limited fields of view, robots can’t always identify, or navigate past, objects around them. Robots also have cognitive limitations. Moving around an office space, for example, a robot could become stumped by rearranged desks, or a new cubicle divider. Depending on whether it has arms and hands, it may also have trouble opening door knobs — a difficult task for any robot that depends on simple claws.
“Robots can do very little,” Veloso told Wired. “I decided that in order to really have these robots be a part of our environment, they need a symbiotic relationship with humans, and they need to proactively ask for help when they need help.” Thus was born the concept of symbiotic autonomy, the driving principle behind Veloso’s CoBots.
In symbiotic autonomy, robots move through the world by themselves, but if they encounter uncertainties about their location, or if what they’re doing surpasses the threshold of their capabilities, they stop and ask humans for help.
“This has been a little bit of a revolution for us,” Veloso said. “To look at these robots as needing help, capable of asking for help, and doing the rest by themselves.” To date, Veloso’s CoBots have traveled more than 200 km on their own, often aided by the helping hand of a human when they need to press an elevator button, for example.
Veloso’s work has garnered herself and her team numerous recognitions and awards, including an IEEE fellowship and a National Science Foundation Career Award.
“I always liked math. I studied electrical engineering as my undergraduate degree,” Veloso said. Her master’s thesis focused on the use of databases in factory inventory automation, and from this she became “fascinated with artificial intelligence and automation in general.”
After finishing a Ph.D. thesis on automated planning algorithms, Veloso became interested in robots as algorithms that would have to accomplish three things: detect the state of the world, generate a plan to achieve goals, and actually execute planned actions — just like humans do. Now, she’s on the verge of a spectacular breakthrough in robotics that could finally let robots successfully commingle among people in public spaces like offices.
Veloso’s CoBot research was born from her influential work in robot soccer, a field that she helped pioneer with Ph.D. student Peter Stone in the mid-1990s. Robot soccer — yes, it’s just what it sounds like — is an example of a very dynamic multi-robot system. Since 1997, its competitive league, Robocup, has been holding an annual championship for teams hailing from around the globe; the 16th annual competition was held at the end of June. The Robocup now has multiple divisions and leagues organized by robot size, ranging from 150 mm tall to the life-sized Humanoid League.
The participating robots have changed dramatically since the early days of Robocup. “Even up until 2002, they could hardly stand and kick the ball. All of our research was on having the biped humanoid machines lift up one foot and kick the ball — and 90 percent of the time they’d fall down,” Veloso says. “Now complete games are run between these humanoid robots. They run on the field, they move fast, they kick, they stand up by themselves. It’s all another story.”
The people behind Robocup have a provocative goal for their robot players, and it’s meant to drive research and innovation: They want their bipeds to beat human soccer teams by 2050. “Of course we are not there yet at all,” Veloso says with a laugh, “but it drives both the development of humanoid platforms that can run on two feet, and lots of other things we’re trying to improve over time.”
Indeed, playing soccer is just a clever exercise for these machines. Veloso and her team are much more interested in the greater field of robotics, and what a kick-ass, football-playing robot can do in the service of humankind. The physical mechanics, AI algorithms, and teamwork employed by successful robot soccer players translates very well to robots used in search-and-rescue missions, as well as service robots for the home or office, like Veloso’s CoBots.
In fact, in the robot soccer world, a lot of the research has focused on complementing what a single robot can (and can’t) do in relation to its team members. For example, if a particular class of robot can only see directly in front of itself, it can be paired with teammates that have a much wider field of vision. The eagle-eyed robots can then wirelessly communicate with the poor-sighted robot, and “magically” the single robot, with all its limitations, becomes empowered. “The reason why I came up with symbiotic autonomy was exactly because I started looking at these robots performing tasks and services for humans as part of a team,” Veloso said.
Veloso believes CoBot-style robots will be the first robots to intermingle with the living, breathing masses. They will help us navigate, act as escorts, allow telepresence, and carry things for us within offices and public spaces, like malls or grocery stores.
Over the next few years, Veloso and her team will be working on giving the CoBots the ability to actively learn semantic labels for locations in an environment; perform natural language-based interaction with humans (like Siri, but via a robot instead of a smartphone); autonomous execution monitoring; and learning by example from human demonstration and human correction.
Currently, our electronics — smartphones, laptops, et cetera — are completely stationary unless we move them ourselves. It will take some time for society to get accustomed to robots roaming among us, performing tasks and meeting deadlines. But that’s what CoBots are already doing within the nine floors of Veloso’s CMU building. “My CoBot project is all about the mobility of these thinking machines,” Veloso says. “These robots are not science fiction.”