Agent-Based Support for Balancing teleoperation and Autonomy in Urban Search and Rescue

Today's artificial intelligence technology is insufficient to support autonomous task performance in complex domains such as urban search and rescue, forcing extensive reliance on human teleoperation of robots. This, however, also has problems: humans quickly suffer from cognitive overload and have difficulties in constructing a representation of the space around a remotely placed robot. In this paper we describe an approach to multirobot control for such environments that focuses on combining the limited abilities of modern autonomous control systems together with human control. At the centre of this approach is a pair of software agents running on each robot: one to recognize problems in the environment from the perspective of a robot, and another to mediate the interaction between a robot and a human controller. The intent of this approach is to allow a human to better control a team of robots, being interrupted only when the situation demands. We describe the implementation of this approach using simulated Pioneer robots, and evaluate the approach in comparison to autonomous and teleoperated mobile robots in a rescue domain.