Interactive task planning under uncertainty and goal changes

Abstract

This paper describes a task planner designed for service robots. Task planning for service robots differs from traditional planning in the following ways: (i) it requires more extensive interactions between the robot and the human. This interaction may convey changed objectives and goals, and advice or directions to perform tasks in a preferred way, (ii) the presence of multiple agents, such as other robots and humans in the work environment. This makes the environment dynamic, since these agents could make changes even as the robot is executing a predetermined plan, and (iii) lack of structure in the environment (in contrast to most industrial applications). This makes the task of sensing relevant conditions less reliable and less feasible, so the robot often has to plan with incomplete knowledge of the environment. This requires the task planner for service robots to be interactive, reactive, and possess the ability to dynamically evaluate plans based on their reliability in uncertain conditions. This paper describes a planner which uses a spreading activation mechanism to combine two approaches to planning: (i) a proactive component that generates plans biased toward picking the most reliable actions in a given situation, and (ii) a reactive component that alters action selection based on changes that occur in the dynamic and uncertain environment. A number of examples related to the ISAC system for feeding the handicapped demonstrate the effectiveness of the planning mechanism.