A model cor controlled planning of object manipulation using multiple agents: an overview

摘要

In implementation of intelligent mechanisms such as dexterous mechanical hands, reconfigurable fuxtures or part feeders, it is always required to plan to manipulate object in ordr to gain its proper configuuratioln within the work-space of the manipulating agents. Most of the proposed planners for coordinated control have been based on ad-hoc approaches using search strategies embedded within varous search trees. What has been lacking in these algorithms is a formal framework for analysis, synthesis and evaluation of their performances. This paper presents an overview of application of such possible controlled planner architecture for manipulating objects using multiple agents. The controller is based on discrete-event control theory. The paper extends the application of such theory to the case of multiple agents manipulating a common object. Here for example, a set of events for lifting and rotating an object is described as labeled alphabets. Then it is shown how the manipulation of obiect can be describved as finite-state automation. The set of uncontrollable events is constructed. The supervisor (planner) can then be constructed by first defining the details of its legal behaviors.