This is the first year of a three-year continuing award. The research explores the information requirements of robot tasks, for automating design of sensors from the specification of a robot's task, its actions, and its uncertainty in sensing and control. The sensors provide precisely the information required by the robot to perform its task, despite the sensing and control uncertainties. The key idea is to generate a strategy for the robot task by using a backchaining planner that assumes perfect sensing while taking careful account of control uncertainty. The resulting plan indirectly specifies a sensor that tells the robot when to execute which action. The sensor need not provide perfect information, but only that information required for the plan to function correctly. In contrast with traditional approaches to task planning which use a bottom-up design that focuses on the sensor characteristics and accommodates them, this approach is driven from the top, by questions such as "given a task, what sensor characteristics and tolerances are needed," and "how does sensing depend on action." The research investigates special purpose sensors for several simple assembly tasks, including multiple-peg-in-hole assembly, parts identification, parts orienting, and sorting. The ultimate goal is to develop task-level planners that can translate high-level task specifications into low-level robot commands.
