Development of a social learning mechanism for a humanoid robot

A main purpose of recent humanoid robotic research is to develop a socially interactive robot by providing for a certain degree adaptability and flexibility in order to endow the robot with natural interactions with humans. In this paper, a social learning mechanism is proposed for enabling a humanoid robot to learn social behaviors through imitation. To achieve this goal, a novel imitation algorithm is proposed for transferring human social behaviors into a robot in real time. This approach considers the characteristic of motions for extracting symbolic postures, which consists of changing the points of motion directions. Reinforcement learning is utilized for extracting optimal symbolic postures and for incorporating the divisional cubic spline interpolation for generating a robot’s social behaviors through symbolic postures. In our experiment, we attempt to transfer three social cues: a “pointing gesture,” a gesture for “explaining something attractively,” and a gesture for expressing “I don’t know.” The experimental results confirmed the accuracy of the robot motion generation through the proposed mechanism for transferring natural social behaviors.