Getting large teams of robots to collaborate is the work of Dr Regis Vincent, who envisions applications that include mapping nuclear contamination.
by Dave Waugh
Dr. Regis Vincent, presenting “Autonomy as the Killer App” on Wednesday, November 2, 2011 at the RoboBusiness Leadership Summit in Boston, is senior computer scientist at SRI International, a nonprofit research institute conducting client-sponsored research and development in the sciences and technology. The focus of Vincent’s research at SRI is the collaboration of large teams of robots.
Dr. Vincent spoke with Robotics Trends writer Ellen Muraskin about what he calls “assistive autonomy,” as a preview to his presentation at the RoboBusiness summit.
Robotics Trends: Put “large team of robots” and “autonomy” in the same sentence and I can’t help thinking of a sci-fi, innovation-run-amok disaster. What do you really mean by “autonomy” here?
Vincent: We don’t really want the robot to think for himself. We’re working on what we call “assisted autonomy,” in which I want the robot to do the boring, repetitive tasks it does well, but leave the harder tasks and decisions to me.
For example, say we have a robot that can move and grasp an object. All the movement to that object, the avoiding of obstacles, the extension of the arm, are the boring and automatic parts of the task. Once the object is grasped, however, all the manipulation of that object may be remote controlled by humans. Then once that task is performed, I want automation to take over again. I want to tell that robot “go and dock yourself” and have that robot find its way back to its docking station and recharge without my having to worry how it does that.
Robotics Trends: What has happened with the Centibots 100-robot project since it has evolved into the Teambotica toolkit? [In the Centibots project, a lead team of mapping robots surveyed an area while building and sharing a distributed map. They were followed by a second wave of tracking robots that configured themselves to efficiently search for an object of interest within that area, sensed and tracked intruders, and shared information among themselves and with a command center.]
Vincent: We’re seeing what other kinds of information a robot can collect and correlate with precise location. In the nuclear environment, for example—a great area of interest since March—robots can to map out areas of radioactive contamination. Or think of gas leaks. Any application in which you want to overlay the position with some sensor reading is the idea.
Robotics Trends: Has this been commercialized yet?
Vincent: Yes. We have robotic mapping software called Karto, available for download on our website. We have about 40 clients who use that to develop robotic applications on our technology.
There is an open source version of Karto, hosted by our friends at Willow Garage. There is also a commercial version with more features, which is closed source. It’s used for higher-end robotic systems.
Robotics Trends: How do you coordinate the “autonomous” movement of so many robots?
Vincent: We just implemented a route system where the robots travel in one direction on the right side of the wall, and the other [robots travel] in the other direction on their right. You treat those as highways of travel. Then you need to coordinate points on the map [where those lanes intersect], and those points have a [virtual] traffic light. You coordinate that. That’s one easy way of avoiding collision.
The nice part about robots is that they’re very predictable. They follow the rules of the road all the time. The problem is humans. If a robot encounters a human, it is programmed to stop and let the human pass before resuming its task.
Robotics Trends: Does the human system administrator have a sort of flight traffic controller interface? What control does he have?
Vincent: He has a God’s-eye view of all the robots’ locations and what they’re doing. He could ultimately remote control them, but with higher robots that’s unlikely to be necessary unless something drastic happens, like if a robot becomes disabled and stops responding.
It works more like asset management. The human has a list of tasks he wants done, and he takes a bunch of robot assets and determines, “OK, you guys have to clear that portion of the building.” How that happens, where they should go, is not the human’s responsibility. It’s the computer’s responsibility to orchestrate that. The human just assigns his chosen number of robots to each task.
Robotics Trends: What kind of computer is running this multi-robot negotiation program?
Vincent: It’s mostly a single laptop. It doesn’t need much compute power. The hardest task is getting all the info to that computer reliably.
