We propose a flexible interactive approach to knowledge based model construction. The domain knowledge is represented in the form of multilevel influence diagrams. At each level, an influence diagram may be generated and the influence diagrams between the levels are related through a series of operations. We defined eight such operations and show that these are sound as well as sufficient to construct any target influence diagram in any practical situation. We have developed an algorithm for model construction using our multilevel influence diagram representation.
Qualitative simulators can produce common sense abstractions of complex behaviors, however, they can also produce an intractable explosion of meaningless behaviors because they attempt to combinatorially order uncorrelated events. People who build brick walls obtain bricks, cement, and tools, and proceed to lay the wall. They don't worry about whether they obtain bricks then tools then cement, or cement, then tools, then bricks, or cement, then bricks, then tools, or cement and bricks, then tools, etc. Common sense tell them that the order in which the events are completed does not matter, what matters is that the events are completed before the wall is laid. Qualitative simulators fail the common sense challenge when confronted with similar problems. Simulators such as QSIM (Kuipers 1994) attempt to calculate all possible orderings of inherently unordered events which can lead to intractable branching in models of even modest size. This paper presents a representation (L-behavior diagrams), an algorithm (L-filter), and a simulator (LSIM) which manages this complexity.
This article describes REBA, a knowledge representation and reasoning architecture for robots that is based on tightly-coupled transition diagrams of the domain at two different levels of granularity. An action language is extended to support non-boolean fluents and non-deterministic causal laws, and used to describe the domain's transition diagrams, with the fine-resolution transition diagram being defined as a refinement of the coarse-resolution transition diagram. The coarse-resolution system description, and a history that includes prioritized defaults, are translated into an Answer Set Prolog (ASP) program. For any given goal, inference in the ASP program provides a plan of abstract actions. To implement each such abstract action, the robot automatically zooms to the part of the fine-resolution transition diagram relevant to this action.
Sophisticated cooperative reasoning among agents requires that the agents be able to share models of their reasoning processes. Developing intelligent systems costeffectively necessitates that components be reused. In order to facilitate sharing and reusing knowledge among distributed (knowledge-based) applications, we have been developing a canonical representation for acquiring and transmitting semantics. This representation will include both a scheme for representing semantics and a hierarchy of concepts used to describe predicates. We have been developing and formalizing a hierarchy describing knowledge to facilitate (1) the specification knowledge and assumptions employed by a system, (2) transferring knowledge among agents (applications) in system and (3) for resolving conflicts among these agents (Figure 1). We are utilizing derivatives of object diagrams [Rumbaugh91], semantic networks [CQ69], and conceptual dependencies [SR74] to describe the fundamental concepts which underlay various algorithms and knowledge representations. By formally describing higher level concepts via these fundamental concepts, we intend to reason about the semantics of and translate knowledge among applications employing different knowledge representations. This hierarchy of concepts will be validated by performing analysis on a complex domain (probably plastics manufacturing) to ensure that the formalisms used to describe elements in the hierarchy are sufficient.
An engineer with a circuit diagram has no need of a knowledge base with explicit rules. The ease and absolute certainty with which engineers can draw conclusions from such diagrams made and makes it very tempting to employ such graphical representations for the knowledge and for the results of the configuring process. When circuit diagrams are available, why should our computer-based configuration tools still have need of separately maintained rules? Werner E. Juengst Motivation Manufacturers of complex products have no real choice but use some form of knowledge representation for describing their product offering and determining the configuration for their products. So they employ the existing inefficient rule-based technologies in spite of the large maintenance effort and accept the huge price tag and the shortcomings. Examining more closely what manufacturers are willing to undergo today, some preconceptions of what modelbased techniques should strive for were found to be astray. Most important is that manufacturers are prepared to describe the product feature combinations they want to offer in rules of propositionalogic, i.e. completely, extensively and explicitly, whereas the preconception of researchers is to treat configuration space as unlimited, and construct the solution incrementally from a minimized knowledge base. Another observation is that many manufacturers today create and maintain, for every component (or for every partial bill-of-material) that might be put into the product, Freightliner Corporation 4747 North Channel Avenue Portland, Oregon 97217 WernerJuengst@Freightliner.com