Deciding Shapes and Motions of a Robot based on Personal Preferences

User preferences for robots differ according to the robot shape and how it moves. This paper proposes a method for generating motions that will be pleasing to a user of a modular robot that can be configured arbitrarily. We propose a preference estimator that uses supervised learning to estimate how well users will like a combination of a motion pattern and robot shape. We also propose a motion selector and a motion generator that select and generate motions in consideration of the balance of the robot's shape and its motion. These methods were tested with a prototype modular robot system and a group of human test subjects who configured the robot as they wished. The proposed motion selector and generator, trained on a preprogrammed set of motion patterns, then presented a ranked set of motions to the subjects. The motion selector proved effective but the motion generator did not. Overall, the proposed system was received favorably by the subjects.