Query Generation for Resolving Ambiguity in User's Command for a Mobile Service Robot

This paper describes a method of generating queries for resolving ambiguities in the user's command in service robotics applications. We deal with bring-me tasks, in which a robot brings a user-specified object from a distant place. A user's command may be ambiguous due to various reasons such as the uncertainty in his/her knowledge of the distant scene and the robot's knowledge. In such a case, the robot compares its recognition result with the command and generates a query for disambiguation. Based on previous VDQG (visual discriminative question generation) work, we develop a method for query generation using the concept of attribute contrast with the attribute categorization. We verified our method by comparing the user and the generated queries. We also implemented a robotic system, as a proof-of-concept, that can interact with the user and certainly achieve bring-me tasks.