Inferring human intent in remote-control scenarios

The ability of inferring human intent can be instrumental in many human-robot interaction scenarios. Inferred intentions can be at both tactical and strategic levels. As remote operation becomes more prevalent, complex new interaction scenarios are emerging, e.g., a human remotely controlling a robot which interacts with another robot. The addition of a remotely controlled robot present further challenges to the already challenging task of inferring human intent. These stem from the capabilities and characteristic of the remotely controlled robot and the distributed structure of the system. We have constructed a robotic testbed for examining inference of human intent in remote control scenarios. The testbed is a form of the table-soccer game where two robots facing each other hit a ball placed within a game court. One robot is controlled autonomously using machine vision and learning algorithms, and the other is controlled by a remote human operator. The game was modeled based on several similar games, where a human plays with a robot (air-hockey, table-tennis). There are two modes of the game which differ based on target: cooperation and competition. In the cooperation mode the target of the players is to achieve the highest number of consecutive passes. In the competition mode the target of the players is to score a goal in the other player's half of the court. The autonomous robot must infer the direction in which the controlled robot will send the ball (tactical) and the game mode which the human has selected (strategic). The paper describes the construction of the system components and their initial testing.