finin@PRC.Unisys.COM (06/24/88)
Date: Thu, 23 Jun 88 13:27 EDT From: finin@PRC.Unisys.COM To: ailist@ai.ai.mit.edu Subject: Unisys AI seminar: The Causal Simulation of Ordinary and Intermittent Mechanical Devices AI SEMINAR UNISYS PAOLI RESEARCH CENTER The Causal Simulation of Ordinary and Intermittent Mechanical Devices Pearl Pu University of Pennsylvania The causal simulation of physical devices is an important area in the field of commonsense reasoning of the everyday physical world. When a human expert describes the way a physical device works, for example a pendulum clock, he or she uses commonsense knowledge of physics and mathematics. To make computers to do likewise, we must first construct a knowledge representation scheme that captures commonsense knowledge, and supports causal simulation. Mechanical systems, especially those that exhibit intermittent motions, provide a good basis for the investigation of behavioral reasoning issues. Our key observation is that the spatial configuration of mechanical devices changes periodically. So far only simple links or conduits have been used to model the connection between a pair of objects in the field. We offer a solution which uses a separate representational entity, called the connection frame, to model the spatial relationships between a pair of objects and how those relationships achieve force or velocity propagation. The connection representation is assumed supplied as part of the design knowledge of the mechanism, though it could be just as readily computed by other spatial connection determination methods. In this talk, I describe a framework constructed to simulate the behaviors of regular and intermittent mechanical systems, with an emphasis on force and velocity propagation reasoning. In general, it appears that continuous motion can usually be modeled by velocity propagation while intermittent motion is best approached by force propagation. The second part of the talk, I discuss a simulation system which attempts to reason about how the physical devices work by simulating the devices qualitatively, mimicing the way people perform such a task. The simulation algorithm will be outlined. Several examples analyzed with the model include dozens of generic objects and connections, a two-gear device, a spring-driven cam mechanism, and a pendulum clock. Currently the simulation is being implemented on the Symbolics Lisp machine in Flavors, which is an object-oriented language. Some of the implementation issues will be discussed as well. 2:00 pm Wednesday, June 29 BIC Conference room Unisys Paoli Research Center Route 252 and Central Ave. Paoli PA 19311 -- non-Unisys visitors who are interested in attending should -- -- send email to finin@prc.unisys.com or call 215-648-7446 --