Plan Generation and Execution for Robotics

A robot device operates by performing a sequence of actions, drawn from a reasonably small repertoire of action types. Industrial robots commercially available today are typically configured to perform a fixed sequence of actions, cued by some visual or contact stimulus. The action sequence is not subject to any significant alteration -- it is performed identically each time the stimulus is presented. Such robots can be called reflex robot systems, since they make a rather inflexible response to a fixed stimulus. The applicability of such systems is limited. Truly flexible automation must make use of significant sensory feedback to respond appropriately to each new stimulus. Thus, a worthwhile goal for robotics research is to develop the capabilities needed to create what one may call instrumental robot systems. The term instrumental is used to suggest, by analogy with classical psychology, the mediation of explicit goals and deliberately initiated actions (i.e., goal-oriented behavior) in the performance of the robot system. The development of such instrumental robots will require extensions of current capabilities in many areas, including control systems, mechanics, and sensors. At the core of the capabilities to be developed lies fundamental research in problem solving and the monitoring of plan execution. This paper discusses a number of problems that are critically in need of further research in support of advanced robotics, including dealing with time, planning for parallel execution, planning for information gathering, planning for planning, learning, interactive planning, dynamic plan repair, and distributed robotics. Presented at the Workshop on Robotics Research held in Newport, RI, on 15-17 Apr 1980. Technical Note 209. SRI Projects 1349 and 8871. Sponsored in part by the Office of Naval Research (ONR).