Design of Representation Systems for Autonomous Agents

摘要

This work proposes a design methodology for creating representation systems for robots that operate in dynamic, uncertain domains. In order to be efficient and effective in these domains, robot control must be achieved through a tight coupling of sensors and effectors and this constrains the design of representation. My methodology attempts to guide the designer in creating representations that are effective for robots using these control systems. The methodology assists the designer in analyzing the robot’s task, capabilities and environment to answer the questions of what to represent, how to structure that representation and how to keep that representation consistent with a changing environment. The efficacy and validity of the methodology is documented through several agent designs that were successfully carried through to implementation. The key feature of these agents is the action-oriented portion of their architectures and the systems of representation they employ.