We have been developing synthetic methodology for understanding the human-humanoid interaction and building an environment where the humanoid robot learns from a long-term and continuous interaction. Along the sketched scenario, we built a humanoid robot Genie which aims at operating for 24 hours and 365 days to learn from interaction with the surrounding environment including humans.

It is considered that the key issue for future cognitive machines is the internal representation of spatio-temporal patterns as well as external representation. In addition, the role of motivation of robots or machines is really important and significant point of view for understanding action-oriented perception. When the robot needs to explain why a particular move was made, the robot needs to explain its motivation. The reasoning is therefore much tied to the task achievement. For example, we humans cannot recognize an object without biases, such as pre-knowledge of the object and environment.

We are currently developing another synthetic methodology for understanding dynamics of objects by a humanoid robot in a psychological framework. The action selection mechanism of the humanoid robot is implemented based on the selection probability consisting of several internal variables that takes insights from human's motivation mechanism. This makes the robot classify its action space by sensory feedback caused by its own action. The system generates a priority to actions which are expected to cause distinguishing sensory patterns. The action selection probability allows the robot to explore unknown spaces for understanding dynamics of objects in a real environment. The preliminary experimental results show that the system allows the robot to build knowledge about the common object movement by repeating interaction with several different types of object, and also to predict the movement of unknown objects.