Planning and distributed planning based on objects and relations with goal-directed dynamic plan revision.

摘要

The research in this thesis involves three areas in the field of artificial intelligence: planning in general, distributed planning and problem solving, and dynamic plan revision.; One of the most important and difficult problems in planning is to determine the effects of an action taken in the world. Due to its inherent complexity and dynamic nature, no coherent theory has been found as yet. A new approach for reasoning about actions is introduced, which is supported by the concepts of objects and relations. This approach makes use of the way human beings view the world, facilitates world modelling and enables the system to determine the effects of actions. Indirect effects, caused by either physical or logical dependency relations, can also be handled appropriately. To deal with the problem of combinatorial explosion in regard to the number of relations a system has to maintain, incremental relation management is proposed.; In real-life applications, one must always be prepared for plans to fail. Complete replanning every time an unexpected change occurs is not a practical solution. Goal-directed dynamic plan revision based on the concept of goal hierarchies has been addressed in this dissertation, which modifies only the plan steps related to the violated goals. Different types of unexpected events have been examined and the types of events causing "benefit" have also been recognized and handled. The notion of system expectations has been discussed, which lets a system have an active attitude toward feedbacks from the environment and react to them properly.; Distributed planning and problem solving is a new and promising research area. An analysis on the cooperation at different phases of distributed problem solving processes is conducted. Two new issues have been identified and studied: the timing problem and the network perception problem both of which contribute to coherent network behavior. The "perceive-plan-act" loop has been introduced. The coordination between nodes has been improved and global coherence enhanced.; A system, based on the research results obtained, has been designed and implemented. The domain of Distributed Manufacturing Operations has been chosen as the testing environment, and experimental results have been shown.