Robot Learning
Full Details
This project focuses on the training of robots to learn new actions and to communicate about their action capabilities. A teacher robot (the agent on your left in the movie) first gives a visual demonstrator to the learner robot (agent on the right) on how to perform actions such as "close_left_elbow" or "raise_right_arm". It also teaches the learner how to name these actions. Subsequently, the learner is able to autonomously acquire new action capabilities (e.g. "grabbing") simply by listening to a linguistic instruction explaining that "grabbing consists of close_left_elbow and close_right_elbow". This instruction can be given by the teacher robot, or by a human instructor speaking directly to the learer robots. The main theoretical and technological implication of such a project is that robots can acquire a language through direct interaction with their physical and social world. This research has been sponsored by the US Air Force Research Laboratory, and has been recently featured in the "New Scientists" and the "Professional Engineering" magazines.