Simulating the Knowledge Environment for Autonomous Construction Robot Agents
Future intelligent construction machines must harness considerable knowledge to plan and control autonomous tasks in spite of the fact that they will be limited in their own pre-defined knowledge. This paper first describes the need to discover and formulate a general core of theory and software for such machines so that they can access and communicate with knowledge sources in their environment. It next describes current research to simulate characteristics of the knowledge environment for
... t agents. An example will illustrate an early implementation effort that uses object-oriented programming for such a simulation. This research will provide a theoretical base for the knowledge environment to sustain automation research for autonomous robots working in real and very challenging field conditions. INTRODUCTION AND BACKGROUND In future construction field environments, intelligent machines, like their human counterparts, will need to harness considerable knowledge to plan and control autonomous tasks in spite of the fact that they will be limited in their own knowledge and abilities. However, no unifying theory and few guidelines exist for defining and communicating knowledge about designs and field operations in a way that can effectively be utilized by such machines. At Stanford and elsewhere, researchers are working toward such a theory in order to support the work of others on practical applications of robots in field conditions [Paulson 85, Paulson et al 89]. Some of the research focuses on cognitive aspects of future machines to endow them with common modes of "thinking" and communicating-in effect a common culture-so that they can work together and with humans in groups. The scope of research needed to build theories and software to support construction robotics is vast. In general, researchers need to develop machine agents having enhanced abilities to work well in relatively unstructured and fast-changing environments. Each step in this research should lead toward a general architecture handling the knowledge an agent needs to function productively in a knowledge environment. The resulting software could then be extended by developers of applications-oriented robots to handle particular areas of expertise, whether in managing other machines or in doing specific physical tasks. This basic research would provide a platform for more rapid integration of robotic systems. Figure 1 shows a broad conceptual view of the construction knowledge environment. It illustrates the organizational context in which the robots might be working, the interfaces to computer-aided design (CAD) databases and reasoning, interactions with other field agents-both human and machine-and interfaces to knowledge sources in the world beyond the field. To establish such an environment with robotic hardware is not feasible now. Even if one could afford them, robots with sufficient flexibility and computing power for diverse operations do not yet exist.