This iCub robot has now learned archery. Only things that was taught for this robot was how to hold a bow and then shoot an arrow. Using this, it improved its accuracy and aiming skill all by itself. Later it became so successful that it started to hit bulls-eye after 8 trials.
The distance between the humanoid robot iCub and the archery target was 3.5 meters. The grip of the bow held by iCub was set manually. Robot's index finger is used to pull/release the string and finally shoot the arrow. There was an algorithm written which helped iCub to remember the correct posture of its left arm and the orientation of the right arm. The algorithm used here is called as ARCHER ( Augmented Reward Chained Regression).
We all know that the goal in archery is to hit the bulls-eye. Each time when iCub shoots the arrow, it gains experience and self tunes itself. In the next attempt it modulates and co-ordinates the movements of its hands accordingly. The movement of the robot's arms is controlled by an inverse kinematics controller. Whenever an arrow is shot, a picture of the target is taken by a camera. An image recognition system has been developed based on Gaussian Mixture Models which helps in recognizing where the arrow head has hit on the target. The data or the information that is collected is then sent as an feedback for the algorithm. Thereby the accuracy of the shot is increased in the next round.
ARCHER algorithm was developed by Dr. Petar Kormushev and his colleagues who belong to Italian Institute of Technology (IIT). In the month of December, Humanoids 2010 conference will be taking place at Nashville TN, US where the results of this research will be presented.
