Generating Abstraction Hierarchies presents a completely automated approach to generating abstractions for problem solving. The abstractions are generated using a tractable, domain-independent algorithm whose only inputs are the definition of a problem space and the problem to be solved and whose output is an abstraction hierarchy that is tailored to the particular problem. The algorithm generates abstraction hierarchies that satisfy the `ordered monotonicity' property, which guarantees that the structure of an abstract solution is not changed in the process of refining it. An abstraction hierarchy with this property allows a problem to be decomposed such that the solution in an abstract space can be held invariant while the remaining parts of a problem are solved. The algorithm for generating abstractions is implemented in a system called ALPINE, which generates abstractions for a hierarchical version of the PRODIGY problem solver. Generating Abstraction Hierarchies formally defines this hierarchical problem solving method, shows that under certain assumptions this method can reduce the size of a search space from exponential to linear in the solution size, and describes the implementation of this method in PRODIGY. The abstractions generated by ALPINE are tested in multiple domains on large problem sets and are shown to produce shorter solutions with significantly less search than problem solving without using abstraction. Generating Abstraction Hierarchies will be of interest to researchers in machine learning, planning and problem reformation.

This book focuses on the performance of mobile robots through the use of multi-hierarchical symbolic representations of the environment. To perform deliberative actions, a robot must possess some symbolic representation of its workspace, but representations of real environments can become so large that they must be conveniently arranged to facilitate and, in some cases, make possible their use. Practical solutions tested on real robots, for example a robotic wheelchair, are provided.

Artificial Intelligence Planning Systems documents the proceedings of the First International Conference on AI Planning Systems held in College Park, Maryland on June 15-17, 1992. This book discusses the abstract probabilistic modeling of action; building symbolic primitives with continuous control routines; and systematic adaptation for case-based planning. The analysis of ABSTRIPS; conditional nonlinear planning; and building plans to monitor and exploit open-loop and closed-loop dynamics are also elaborated. This text likewise covers the modular utility representation for decision-theoretic planning; reaction and reflection in tetris; and planning in intelligent sensor fusion. Other topics include the resource-bounded adaptive agent, critical look at Knoblock's hierarchy mechanism, and traffic laws for mobile robots. This publication is beneficial to students and researchers conducting work on AI planning systems.

"The central fact is that we are planning agents." (M. Bratman, Intentions, Plans, and Practical Reasoning, 1987, p. 2) Recent arguments to the contrary notwithstanding, it seems to be the case that people-the best exemplars of general intelligence that we have to date do a lot of planning. It is therefore not surprising that modeling the planning process has always been a central part of the Artificial Intelligence enterprise. Reasonable behavior in complex environments requires the ability to consider what actions one should take, in order to achieve (some of) what one wants and that, in a nutshell, is what AI planning systems attempt to do. Indeed, the basic description of a plan generation algorithm has remained constant for nearly three decades: given a desciption of an initial state I, a goal state G, and a set of action types, find a sequence S of instantiated actions such that when S is executed instate I, G is guaranteed as a result. Working out the details of this class of algorithms, and making the elabora tions necessary for them to be effective in real environments, have proven to be bigger tasks than one might have imagined.

AAAI proceedings describe innovative concepts, techniques, perspectives, and observations that present promising research directions in artificial intelligence.AI and Education. Automated Reasoning: automatic programming, planning and scheduling, rule-based reasoning, search, theorem proving, uncertainty, truth-maintenance systems, constraint-based systems. Cognitive Modeling. Commonsense Reasoning: qualitative reasoning, design, diagnosis, simulation. Impacts of AI Technology: organizational, economic, and social implications. Knowledge Acquisition and Expert System Design Methodologies: techniques for designing expert systems and acquiring domain knowledge. Knowledge Representation: knowledge-representation systems, inheritance, nonmonotonic logic, nonstandard logics, temporal reasoning. Machine Architectures and Computer Languages for AI. Machine Learning. Natural Language: generation and understanding; syntax, speech, dialogue. Perception and Signal Understanding: vision. Philosophical Foundations. Robotics. User Interfaces.