Nico recognizes its arm's location and orientation down to accuracy of 2 centimetres in any dimension
NICO spends a lot of time looking in the mirror. But it's not mere vanity - Nico is a humanoid robot that can recognise its reflection - a step on the path towards true self-awareness.
In fact, Nico can identify almost exactly where its arm is in space based on the mirror image. Justin Hart and Brian Scassellati at Yale University have taught Nico to recognise the arm's location and orientation down to accuracy of 2 centimetres in any dimension. It is a feat of spatial reasoning that no robot has ever accomplished before.
Motion sensors in Nico's arms relay information when his arm is moving to his software system. The software compares the images received by the camera eye with the movement of the arm, and when they align, there is a good probability of the image being Nico himself. If not, the software determines a higher probability that the image is someone or something else.
Nico is the centrepiece of a unique experiment to see whether a robot can tackle a classic test of self-awareness called the mirror test. What does it take to pass the test? An animal (usually) has to recognise that a mark on the body it sees in the mirror is in fact on its own body. Only dolphins, orcas, elephants, magpies, humans and a few other apes have passed the test so far.
Precise recognition of where its body is in space will be key if Nico is to get to grips with the mirror test, which by its nature is performed in 3D. Before it does, though, the robot will need to learn more about itself. The team plan to teach Nico how to recognise where its torso and head are, what shape they are, and their colour and texture so it can see and react to the mark on its body. Nico already understands how to connect movement of its limb to motion in its reflection, another important skill it achieved in an experiment in 2007.
"What excites me is that the robot has learned a model of itself, and is using it to interpret information from the mirror," says Hart. He and Scassellati presented the work last month at the Conference on Artificial Intelligence in Toronto, Canada.
Mary-Anne Williams of the University of Technology Sydney, Australia, points out that robotic self-awareness is crucial if robots are ever going to work safely alongside humans. "Many robots today not only do not recognise themselves in a mirror, but do not recognise their own body parts directly," she says. "For example, a robot may be able to look downwards and see its feet but not recognise them as its own."
Self-awareness is a basic social skill and without it robots will struggle to interact with people effectively, Williams adds.