Showing papers by "Qiang Yang published in 1994"

A theory of conflict resolution in planning

Abstract: Conflict resolution in planning is the process of constraining a plan to remove harmful interactions that threaten its correctness. It has been a major contributing factor to the complexity of classical planning systems. Traditional planning methods have dealt with the problem of conflict resolution in a local and incremental manner, by considering and resolving conflicts individually. This paper presents a theory of conflict resolution that supports a global consideration of conflicts. The theory enables one to formally represent, reason about and resolve conflicts using an extended framework of constraint satisfaction. The computational advantage of the theory stems from its ability to remove inconsistencies early in a search process, to detect dead ends with low computational overhead, to remove redundancies in a search space, and to guide the search by providing an intelligent order in which to resolve conflicts. The paper also presents empirical results showing the utilities of the theory, by investigating the characteristics of problem domains where the theory is expected to work well, and the types of planning systems for which the theory can offer a marked computational advantage.

Delaying variable binding commitments in planning

Abstract: One of the problems with many partial-order planners is their eager commitment to variable bindings. This is contrary to their control decision in delayed-commitment of operator orderings. In this paper we present an extension to the classical partial-order planners by associating every variable with a finite domain of values. The extended planner applies constraint satisfaction routines to check for the consistency of variable bindings. With the extended planner, we perform a set of experiments to show that it is able to reduce the amount of backtracking in domains where few variable instantiations can lead to final solutions. We also investigate the frequency of CSP application to determine how often should constraint checking be performed in order to yield the best performance.

Evaluating the Trade-Offs in Partial-Order Planning Algorithms

Abstract: : Most practical partial-order planning systems employ some form of goal protection. However, it is not clear from previous work what the tradeoffs are between the different goal protection strategies. Is it better to protect against all threats to a subgoal, some threats, or no threats at all? In this paper, we consider three well-known planning algorithms, SNLP, NONLIN, and TWEAK. Each algorithm makes use of a different goal protection strategy. Through a comparison of the three algorithms, we provide a detailed analysis of different goal protection methods, in order to identify the facts that determine the performance of the systems. The analysis clearly shows that the relative performance of the different goal-protection methods used by the systems, depends on the characteristics of the problems being solved. One of the main determining factors of performance is the ratio of the number of negative threats to the number of positive threats. We present an artificial domain where we can control this ratio and show that in fact the planners show radically different performance as the ratio is varied. The implication of this result for someone implementing a planning system is that the most appropriate algorithm will depend on the types of problems to be solved by the planner. Partial-order planning, Goal protection, Algorithms, TWEAK, SNLP, NONLIN

An evaluation of the temporal coherence heuristic in partial-order planning

Abstract: This paper presents an evaluation of a heuristic for partial-order planning, known as temporal coherence. The temporal coherence heuristic was proposed by Drummond and Currie as a method to improve the efficiency of partial-order planning without losing the ability to find a solution (i.e., completeness). It works by using a set of domain constraints to prune away plans that do not “make sense,” or are temporally incoherent. Our analysis shows that, while intuitively appealing, temporal coherence can only be applied to a very specific implementation of a partial-order planner and still maintain completeness. Furthermore, the heuristic does not always improve planning efficiency; in some cases, its application can actually degrade the efficiency of planning dramatically. To understand when the heuristic will work well, we conducted complexity analysis and empirical tests. Our results show that temporal coherence works well when strong domain constraints exist that significantly reduce the search space, when the number of subgoals is small, when the plan size is not too large, and when it is inexpensive to check each domain constraint.

Search Reduction in Planning with Primary Effects

Abstract: The use of primary effects in planning is an effective approach for reducing search costs, closely related to abstraction planning. However, there has been little analysis of planning with primary effects and few experimental results. In this paper we demonstrate the connection between primary effects and abstraction hierarchies and provide analytical and empirical results on the efficiency of planning with primary effects. First, we show how the relationship between primary effects and abstraction can be used for automatically selecting primary effects and generating good abstraction hierarchies. Then we analytically demonstrate that the use of primary effects may lead to an exponential reduction of the running time and identify factors that influence the efficiency of planning with primary effects. Finally, we describe experiments that confirm our analysis and demonstrate the practical efficiency of using primary effects.

A framework for automatic problem decomposition in planning

Abstract: An intelligent problem solver must be able to decompose a complex problem into simpler parts. A decomposition algorithm would not only be beneficial for traditional subgoal-oriented planning systems but also support distributed, multi-agent planners. In this paper, we present an algorithm for automatic problem decomposition. Given a domain description with a number of objects to be manipulated, our method constructs subspaces complete with subproblem descriptions and operators, and solves the subproblems concurrently. The solutions in individual subspaces are combined using a constraint satisfaction algorithm. The effectiveness of the approach is guaranteed by our careful analysis of the interactions among different subspaces. The results presented in this paper support parallel, distributed and multi-agent planning systems.