An Analysis of Sensor-Based Task Planning

We present a planner which can plan to perform sensor operations to allow an agent to gather the information necessary to complete planning and achieve its goals in the face of missing or uncertain environmental information. We have identiied and addressed two of the chief problems associated with interleaving execution with planning. The rst is that early execution of actions may interfere with, or even preclude the achievement of goals not yet considered. The second is that data needed for planning may be incorrect or uncertain even after they are sensed from the environment. We cast the problem as one of choosing between various sensing policies. For each piece of unknown or uncertain information, the planner must decide on the best way to obtain it in order to minimize the execution cost or provide itself with the greatest chance of success overall. We provide an analysis of the factors that innuence plan quality, and suggest methods which use these factors to evaluate and rationally choose among various sensing policies.