Reactive Query Policies: A Formalism for Planning with Volatile External Information Tsz-Chiu Au Department of Computer Science University of Maryland College Park, MD 20742 Email: chiu@cs.umd.edu Dana Nau Department of Computer Science University of Maryland College Park, MD 20742 Email: nau@cs.umd.edu Abstract— To generate plans for collecting data for data mining, an important problem is information volatility during planning: the information needed by the planning system may change or expire during the planning process, as changes occur in the data being collected. In such situations, the planning system faces two challenges: how to generate plans despite these changes, and how to guarantee that a plan returned by the planner will remain valid for some period of time after the planning ends. The focus of our work is to address both of the above challenges. In particular, we provide: 1) A formalism for reactive query policies, a class of strategies for deciding when to reissue queries for information that has changed during the planning process. This class includes all query management strategies that have yet been developed. 2) A new reactive query policy called the presumptive strategy. In our experiments, the presumptive strategy ran exponen- tially faster than the lazy strategy, the best previously known query management strategy. In the hardest set of problems we tested, the presumptive strategy took 4.7% as much time and generated 6.9% as many queries as the lazy strategy. I. I NTRODUCTION One difficulty with integrating planning and data mining is that the results of data collection and analysis may influence the planning process itself. To decide how best to plan for subsequent data-collection and analysis efforts, the planner may need to request information from external information sources such as sensors, databases, data analysis programs, web services, and the like, incurring a lag time for receiving the answers. Furthermore, the planning activity may occur over a period of several hours or even several weeks [1]– [4]. This forces the planner to deal with information volatility during planning: as changes occur in the external world, the information needed by the planning system may change or expire before the planning process completes. AI-planning researchers have not paid much attention to information volatility during planning. Instead, their research has concentrated on static environments in which no changes occur in the world other than the ones caused by executing the planner’s plans. But it is easy to find many practical situations in which information volatility occurs. For instance: • For analyzing large amounts of data in data mining and other applications, grid computing is an increasingly important technique [5]. In grid- and utility-based com- puting applications [6], one might want to reserve com- Fig. 1. A screenshot of an online airline-ticket reservation system. The prices expired while the user was trying to plan other details of the trip. puting resources owned by several different companies though some grid services on the Web to accomplish a computational task, but the availability and the amount of the resources will keep changing. • When collecting and analyzing data from web services, one problem is that the information can change very frequently. In [3] and [4], Kuter et al. describe a domain- specific system that uses AI planning to do web ser- vice composition; this system was explicitly designed to change its plans in response to information volatility. • Readers who have tried to make travel plans will probably recognize the kind of screenshot in Figure 1. Here, the information about an airline flight expired while one of us was trying to plan some other details of a trip. In such environments, how to cope with changes of external information and at the same time generate a plan that can be executed correctly is a big challenge. Most existing planners will not do this correctly unless they are modified. In some cases, the interleaving of planning and execution [7] is a good strategy to deal with volatile information. But if the wrong choice of action can cause a failure that is irrecoverable (or recoverable only at a large cost), then it is necessary for the planning system to reason, while the plan is being generated, about whether the assumptions that are being used to choose an action will still be true when the action is executed. To the best of our knowledge, our previous work [8] was the first to provide a way for planning systems to do such reasoning. A primary limitation of that work was that although