More and more tasks are carried out by robots, or will be soon. Think of care robot AMIGO that can support hospital staff, or do household chores. Today, this type of robot is programmed for each task separately still.

However, robots could be used much more efficiently if they were able to learn from each other, say René van de Molengraft, associate professor at the Department of Mechanical Engineering and director of robot soccer team Tech United, and his German researcher colleague Oliver Zweigle of the University of Stuttgart. Van de Molengraft: “If a personal robot is to become an everyday appliance like a personal computer, and we manage to teach it something it can share with other robots, it could have a major impact. We had to work with that idea.”

Their joint plan evolves into a grant proposal that is subsequently accepted by the European joint program FP7. RoboEarth is given the green light, with Van de Molengraft at the helm of the project. Although it may seem easy on paper to have robots share their knowledge, it still takes a large group of researchers four years to develop a system that can connect robots all over the globe.

Sharing knowledge is nothing new. People do so through the Internet on a daily basis. Still, it’s new for robots. The way knowledge is represented has to be unambiguous, must be stored, and reinterpreted for a robot not to mess things up. And then the robot has to act accordingly. A lot of research has gone into these separate components, but creating a new system that combines them is what makes RoboEarth a high-profile project.

The concept is shaped from several angles. “Each partner in the project has their own expertise that’s essential to make the project work. And with everyone approaching the issue from a different perspective, we can take larger steps”, says Zweigle, who’s working on knowledge storage with the Technical University of Munich.

Van de Molengraft’s group is working on knowledge representation, and researches how to model spaces. The latter is vital to secure objects, and have robots move in a room. The University of Zaragoza is an authority in the field of perception systems and mapping: how does a robot perceive things? The way robots learn is studied at ETH Zurich, where they’re also developing the RoboEarth Cloud Engine, which is used for calculating and interpreting information. The information comes from a database designed in Stuttgart in which everything is stored: knowledge for navigation, performing tasks, object recognition, and a number of software components. In a nutshell, the RoboEarth project involves the creation of a loop from robot to said database, allowing robots to reuse knowledge and learn from each other as a result.

It’s about what goes on inside the robot

And then suddenly, after years of research, Robot Wikipedia is launched. The final demonstration of RoboEarth in mid-January generates a lot of (international) attention. In a reconstructed hospital room – scanned by robot Ari – there’s a patient that would like something to drink. With the help of Ari’s information, care robot AMIGO can serve the patient. The drink ends up on the floor, and there are some minor glitches, but even so the test is an absolute success.

“People want to see robots move around on a day like this, obviously. Still, they have to realize we’re dealing with experimental software. In this case, we’re much more interested in what goes on inside the robot”, Sjoerd van den Dries explains. The Mechanical Engineering doctoral candidate has been working with the RoboEarth project for several years now.

“We’re demonstrating a system design consisting of integrated components that several groups have been working on for some time now. It’s not a product yet. This specific example of serving a drink quite beautifully shows how knowledge is shared. It may sound simple, but someone has to translate that into voltages and motors at one point; we’re working on the tiniest level imaginable. We’ve had to figure out what’s specific and what’s generic to a robot.”

Van de Molengraft: “We’ve developed a prototype; it’s not ready to go into production at all. Of course, we can do much more with this prototype than that of four years ago, but it could easily be ten years before we can present an actual product. In the meantime, we keep pushing forward, of course. We’ve proven that a system like RoboEarth is vital to implement robots usefully and safely, and we now know what requirements such a system should meet.”

The great collaboration between the various partners has turned out to be one of reasons the project is a success. “It’s an amazing multidisciplinary project that elevates everyone because we’re combining our areas of expertise”, Zurich-based PhD candidate Gajan Mohanarajah explains excitedly. “It really is something we can be proud of. We’ve been given the chance to learn from each other, and because of that we did more than develop a complete architecture and algorithms; we’ve increased the project’s applicability.”

To promote collaboration, the project was divided into work clusters, so partners didn’t have to work on separate components on their own. Van den Dries: “Combining a number of disciplines creates a flow of creativity. It’s important you speak each other’s (technical) language, though. And you have to communicate at all times to prevent groups from going in opposite directions. It’s a wonderful setting for PhD students: you get to work with a group of international, highly experienced researchers, which is great for knowledge exchange and a welcome network boost.”

The project involves quite some traveling, but all parties consider that to be a good thing. “We’ve attended innumerable meetings and workshops. The most notable were the integration workshops, for which the whole group would meet somewhere to integrate the separate parts into a single system”, says Zweigle. “And those moment are indispensable, too”, Van den Dries adds. “Talking over the phone or via e-mail is fine if you want to get up to speed quickly, or for solving a specific problem, but real-life discussion are far more animated. Some doctoral candidates involved in the project have spent months with a partner group. Things like that definitely tighten the bonds.”

Van de Molengraft concurs. “All those meetings help create a close-knit team. As a result, we can take larger steps. We’ve spent quite some time on planes, but distances are doable. A European project like this poses a perfect opportunity to get a major study funded. On top of that, all those workshops and demonstrations add greatly to the visibility of our group. We already have a large global network thanks to the RoboCup, but we’ve noticed RoboEarth has drawn us back into the limelight.”

RoboEarth 2.0?

Champagne bottles have popped to celebrate the successful completion of the project, but what will happen next? “Whatever happens, we’ll continue in the same direction”, says Van de Molengraft. “But we’re not yet sure how exactly, or what partners – both from the academic world as well as businesses – will be involved. For RoboEarth, Philips was our industrial partner, and having a major name like that on board is always a good thing. On the other hand, it’s also a good way for companies to develop their knowledge. More and more names, including Google, are realizing they have to collaborate in order to keep a finger in the pie. So over time, the willingness to join such projects will only increase. Within RoboEarth, TU/e had a coordinating role, which was a good way to present ourselves. It’s a plus when submitting new grant proposals.”

Mohanarajah: “There are no real plan for RoboEarth 2.0. but thanks to the long-term cooperation, we all know what the others have in store, so it’s easy to find each other for new projects, maybe on a smaller scale.” And that’s why Van de Molengraft wouldn’t be surprised if in four years’ time there were another major demonstration. “Who knows, it may be in the same hospital setting with AMIGO serving a beverage again. But he’d have many robot colleagues, and he’d be operating in a large room filled with beds. Having RoboEarth as a solid basis, things might just start moving really quickly.”

www.roboearth.org