Using a learning algorithm, Italian researchers taught a child-like humanoid robot archery, even outfitting it with a spectacular headdress to celebrate its new skill.
Petar Kormushev, Sylvain Calinon and Ryo Saegusa of the Italian Institute of Technology developed an algorithm called “Archer,” for Augmented Reward Chained Regression. The iCub robot is taught how to hold the bow and arrow, but then learns by itself how to aim and shoot the arrow so it hits the center of a target.
Watch it learn below.
The researchers say this type of learning algorithm would be preferable to even their own reinforcement learning techniques, which require more input from humans. Kormushev et. al chose archery because of its complexity, requiring coordinated visual processing, the use of both hands and external tools.
The team used an iCub, a small humanoid robot designed to look like a 3-year-old child. It was developed by a consortium of European universities with the goal of mimicking and understanding cognition, according to Technology Review. It has several physical and visual sensors, and “Archer” takes advantage of them to provide more feedback than other learning algorithms, the researchers say.
The team will present their findings with the archery learning algorithm at the Humanoids 2010 conference in December.