Video: the state of the art in robot perception and dexterity
By Loz Blain
September 29, 2009
Don't watch the video after the jump if you've recently seen a Terminator movie - because it's becoming clear that if the robots come after us, there'll be no stopping them. The video shows an incredible array of three-fingered robot hand/eye co-ordination exercises, including throwing and catching, spinning pens, tying knots and dribbling a ping-pong ball. The Ishikawa Komuro laboratory at the University of Tokyo put these videos together to demonstrate the incredibly quick parallel processing they are achieving with a mix of visual and tactile sensory inputs. Astounding stuff. Sarah Connor, you're in deep trouble.
Our own human dexterity is based on the super-quick combination of a number of sensory inputs - we make calculations of trajectory, balance and movement so quickly that we become unaware we're even thinking about them, and act totally by instinct.
Take the act of catching a spinning mobile phone, like the robot in the video manages so successfully. Our eyes are relaying us stereo images of the phone's path through the air. We are simultaneously calculating its arc through the air, its spin on three axes, how these relate to the position of our eyes, what this means compared to the positions of our hands, how to move the hand into the projected trajectory, and when to close our fingers to catch the phone ... among many, many other things.
The fact that our human machinery can do this so instinctively is one of the many magical marvels of evolution - but watching a computer-controlled robot, programmed by humans, learning to piece these parallel tasks together is more amazing still.
Most of the robot's motions are managed so quickly that you've got to watch them in slow-motion to appreciate them. And once you do, you get a sense of the complexity of what the robot is managing to achieve. This stuff is absolutely mind-bending.
And the robot's extreme quickness, in tasks such as the ping-pong ball dribble, is already far exceeding what any human could do - which brings us to the next point: where our human dexterity is limited by the meat and nerves we're given to operate with, robotics are limited only by the technology of the time - which continues to improve in leaps and bounds. And just as this robot is learning to process information in parallel streams from its multiple sensory inputs, robotic technology will improve in parallel streams, with new advances in processing, information storage, sensory inputs and mechanical actuation all feeding in.
Because each task needs to be conceived, designed and programmed by humans right now, development is still fairly slow. But when these machines learn to teach themselves... These are, indeed, amazing times.
Learn more about the Ishikawa Komuro Laboratory and the many abilities of its robotic children over at the Lab's website.