Robots excel in situations where they can be told precisely what to do. Manufacturing robots, for example, serve rigid roles, strictly determined ahead of time and governed by simple rules. Movements are repetitive and exact, timing rarely changes, and errors are coped with in a set of predetermined contingencies.
It wouldn't be accurate to say that the "Justin" robot, a humanoid machine built by the Institute of Robotics and Megatronics at the German Aerospace Center, is intelligent--it's not, in any familiar sense. But it's definitely fair to call it extremely capable. Its most impressive talent is the ability to catch a ball thrown not by a predictable and consistent machine, but by a consistently inconsistent human. Have a look:
What's its trick? Well, it's got a few:
- 3D vision: A pair of 2.0 megapixel cameras sit side-by-side to allow the robot to accurately track object position and distance, giving it something akin to depth perception. In fact, the video recorded by these two cameras could be combined and shown as 3D video--paired lenses are what make 3D cameras work.
- Positional awareness: An inertial measurement unit (IMU) helps the robot keep track of its head orientation and movement. This data is used to put imagery from the cameras into context.
- Smart software: The trajectory of the ball is continuously predicted throughout its flight. Humans (and even dogs) are able to estimate estimate the trajectory of a thrown object with high accuracy without actually performing any explicit calculations; Justin, on the other hand, must process each throw with calculus.
- Touch sensitivity: Catching a ball is mostly a matter of predicting position and timing. Finer movements require the robot to respond to direct physical feedback. A collection of sensors in the machine's hands allow it to make precise movement decisions in situations where the camera doesn't tell the whole story.
Throwing a ball and pouring drinks makes for good footage, but this kind of dynamic response capability has wide practical applications, not just in human interaction but in manufacturing--particularly in complex, highly variable sorting or screening scenarios that currently require a human eye and touch.
Likewise, the same technology that allows Justin to consistently catch a baseball could be used to give a humanoid robot a well-timed and well-positioned handshake. And those sensors in his hand could make sure that its a convincing one--not too limp, not too firm, and most importantly, not too robotic.
Justin will be on display at the ICRA 2011 expo in Shanghai.