It's one thing for mechatronics experts to cluster among themselves, sharing insights on their various disciplines. It's quite another for someone like Eric Gradman, a roboticist at a company called Applied Minds in Glendale, CA to spread the word about his passion for robotics. He spoke at an event called Mindshare in mid-October in Los Angeles, which brings together on a monthly basis an audience as diverse as engineers, technologists, artists and dancers.
Of course, Gradman is not your ordinary roboticist. Sure, he has a master's degree in robotics from the University of Southern California. He believes robotics encompasses mechatronics. "When I talk about why I enjoy robotics so much, I say, when I'm confronted with a problem, only in robotics do I get the choice of solving it in hardware, software or electronics."
But he's also an interactive artist and a circus performer. (Admittedly, roboticists and circus performers both employ a basic understanding of physics in their work).
Nor is Mindshare your ordinary event. According to one of the founders, "We're intrigued by the cross pollination that arises from gathering individuals who might not otherwise interact. It is an energetic mix of open-minded people from architects to programmers to writers to painters to engineers and all others in between."
Clearly, Gradman, Mindshare and mechatronics all share a multidisciplinary view of the world.
Still, Gradman chose a unique perspective with which to discuss robotics with the Mindshare audience. His presentation was titled "The Perils of Robotics: When Good Robots Do Bad Things." When he came up with the title, he wanted to avoid giving a dry presentation about robotics. "I love robotics," he says, "but it may have been that I was having a particularly frustrating week at work, because I got to thinking about the many ways that robots can fail."
So why do good robots do bad things? "Robots can misbehave," says Gradman, who started out his presentation with the example of Kenji Urada, one of the first workers ever to be killed by a robot, "but they can't be evil or stupid. It comes down to a human failure to design around a problem. Robots fail because we don't know how to solve all the problems (in robotics). We don't understand how to make them succeed yet."
Gradman says there are multiple ways robots foul up basic activities they're supposed to understand, including:
Sensing their environment. When robots use lasers as sensors to understand their environment, they can be fooled by reflective materials. The laser can be diffused to a point where it can't recognize an object. This is fine, Gradman says, except very expensive sports cars tend to have exactly the kind of paint job that fool the lasers. "In a parking lot full of cars, the robot will run into the most expensive one," says Gradman. "I think that in the future, to avoid hostile robots, people will wear clothing made of silver lamé and mylar."
Modeling their environment. Gradman works frequently with computer vision, so he's come to understand robots' limitations in this area. In his presentation, he showed examples of a robot accurately tracking the face of a woman who's moving around a room. But he also shows an example of a robot using standard tracking algorithms that is fooled into thinking a smiley face painted on a human hand is another human face.
Planning a course of action. Sometimes engineers ask robots to take an action that requires too much flexibility in too small a space or too short a time, Gradman says. "All of a sudden the robot is asked to do something that's both fast and impossible, and without safeguards you could put people in danger." And sometimes robots just don't do what they're supposed to.
Acting to achieve the goal. And sometimes there's just hardware failure. Gradman included in his presentation video of Honda's Asimo robot ascending a set of stairs, turning to descend a second set of stairs, and toppling forward catastrophically.
Why are robots prone to failure? The answer is simple, in Gradman's eyes. "Robotics is still about human beings putting together robot systems. We become part of the robot for a brief period of time, and idiots can cause more damage through robots than any robot can by itself."
For all his humorous examples of robotics failure, Gradman is highly optimistic about the future of robotics. In September, he attended IROS 2008 (Intelligent Robots and Systems) in Nice, France. "It's very clear from listening to all the talks there that we are rocketing ahead in state-of-the-art sensing, planning and modeling," Gradman says. "Processing power is increasing, so we can build more nuanced algorithms and more sophisticated mathematical techniques to deal with interesting exceptions in data. Sensors are getting better, and our methods of dealing with the sensor data are getting better. Brilliant researchers are making great inroads in all these things, so we're on the path to solving a lot of the problems I talked about."