Balancing Robotic Teleoperation and Autonomy in a Complex and Dynamic Environment

While Artificial Intelligence has been working to produce truly autonomous problem solving agents for many years, the current abilities of such agents are extremely limited. In highly complex, dynamic situations such as disaster rescue, today’s agents simply do not have the ability to perform successfully on their own: the environment is difficult to traverse and even to sense accurately, time is a significant factor, and the dynamic and unpredictable nature of the environment tends to preclude the ability to produce extensive plans for future activity. Because of these and other limitations, robotic agents for environments such as disaster rescue rely strongly on human teleoperation. This too has its limitations: humans become fatigued rapidly, suffer from cognitive overload when they obtain too much sensory information in a short time, and have difficulties in constructing a mental image of the space around a robot given information from its senses (situational awareness). This thesis focuses on combining the limited abilities of an autonomous agent together with human control, in order to produce a teleautonomous system that supports blending the desires of a robot with the wishes of its human controller. The approach I present is intended to allow a human to control a number of robots, being interrupted only when the robots are truly in need, and with the ability to alter the autonomous abilities of the robots for particular contexts. In order to examine the effectiveness of this approach, I develop a simulated domain for disaster rescue using a widely-employed robot simulation tool, and implement this control mode for a set of simulated Pioneer mobile robots. An evaluation of this control mode in comparison to autonomous and teleoperated agents is then presented.