Using constraint programming to model complex plans in an integrated approach for planning and scheduling-猫眼课题宝

使用约束规划以集成的规划和调度方法对复杂计划进行建模

DOI：

发表时间：

期刊：

影响因子：

通讯作者：

中科院分区：

文献类型：

作者：

Antonio Garrido;E. Onaindía;Marlene Arangú

本文的目的是提出一个集成的架构规划和调度（P&S），更具体地说，在这个框架内的约束编程模块（调度器）的作用。其基本思想是有两个独立的模块，一个计划器和一个调度器，共同解决一个P&S问题。在这个合作框架中，我们首先集中我们的研究与复杂的P&S约束作为CSP的计划建模。然后，CSP求解器将该计划连同其所有约束沿着作为输入，并验证这样的计划。本文提出的模型也可以很容易地扩展到解决P&S问题，而不仅仅是验证计划。介绍和动机P&S问题的解决遵循两种不同的观点。在研究方向之一，时间规划方法，其目标是扩展规划，以科普调度能力，即增强规划推理能力，以处理时间和资源（陈，许，和华2005; Gerevini等。2004; Ghallab，Nau，和Traverso 2004）。在另一个方向，计划嵌入调度，解决方案包括将计划功能纳入调度程序（Smith &齐默尔曼2004）。在后一种方法中，起点通常是一组预先计划好的有序活动，每次需要发布、设置或使任何问题组件可用时，都会调用计划。然而，有可能提出一个更普遍和灵活的模型，其中P&S都在问题解决中发挥重要作用，这是一个热门的研究课题。我们使用这样的模型来组合联合收割机规划求解器和CSP求解器（作为调度模块），如图1所示，图1描述了我们的集成架构的结构：输入是问题模型（领域+任何规范语言的问题定义，例如PDDL 3（Gerevini & Long 2006））。此外，我们允许通过将进一步的限制，如定量的时间限制或更复杂的局部条件的行动模型的扩展。整个问题的定义分为两个部分：命题部分，包括那些方面的因果结构的问题和额外的约束条件，其中包括数值约束，偏好和硬约束和额外的约束条件的扩展模型。通过将问题建模分为这两个部分，我们可以使用一个经典的规划器，尽可能简单（在表达性和微积分方面）和有效，来解决问题的命题部分。此外，我们甚至可以使用用户生成的输入计划作为活动序列。重要的是要注意，该计划可以抽象出调度（时间+资源）要求，即计划不需要是可执行的，因为集成模块的目标正是修复给定计划并使其完全可执行。所有的日程安排要求。总而言之，我们可以使用一个纯粹的CNOPS计划器或PDDL计划器或手工定制的计划。在任何情况下，输出都将是手头问题的框架计划或因果结构。显然，我们用于生成计划的方法越先进，计划质量就越好。·造型。在我们的框架中，第二步是问题表述。我们不是编码规划结构（例如规划图）（Kambhampati 2000），而是将计划因果结构和附加约束编码为CSP。因此，本文详细地展示了基于Refanesthet 2005; Vidal & Geffner 2006的作品的具有复杂约束的计划的制定。我们可以将一个完整的计划建模为一个

The objective of this paper is to present an integrated architecture for planning and scheduling (P&S) and, more particularly, the role of a constraint programming module (scheduler) within this framework. The underlying idea is to have two separate modules, a planner and a scheduler, solving jointly a P&S problem. In this collaborative framework, we have firstly focused our research on the modelling of a plan with complex P&S constraints as a CSP. Then a CSP solver takes as input the plan along with all its constraints and validates such a plan. The model presented in this paper can also be easily extended to solve P&S problems rather than just validating plans. Introduction and motivation The resolution of P&S problems has followed two different perspectives. In one of the research directions, the temporal planning approach, the objective is to extend planning to cope with scheduling capabilities, that is augmenting the planning reasoning capabilities in order to handle time and resources (Chen, Hsu, & Wah 2005; Gerevini et al. 2004; Ghallab, Nau, & Traverso 2004). In the other direction, planning embedded into scheduling, the solution consists of including planning capabilities into a scheduler (Smith & Zimmerman 2004). In this latter approach, the starting point is usually a pre-planned set of ordered activities and planning is called each time it is necessary to release, set up or make any problem component available. However, it is possible to come up with a more general and flexible model where both P&S have an important role in the problem solving, which is a hot topic of research. We use such a model to combine a planning solver and a CSP solver (acting as a scheduling module) as presented in Figure 1, which depicts the structure of our integrated architecture: • The input data. The input is the problem model (domain+problem definition in any specification language, e.g. PDDL3 (Gerevini & Long 2006)). Additionally, we allow an extension of the model by incorporating further constraints like quantitative temporal restrictions or more sophisticated local conditions for actions. The overall problem definition is divided into two parts: the propositional part that includes those aspects related to the causal structure of the problem and the additional constraints which comprise the numeric constraints, the preferences and hard constraints and the extra constraints of the extended model. By separating the problem modelling into these two parts we can use a classical planner, as simple (in terms of expressivity and calculus) and efficient as possible, to solve the propositional component of the problem. Moreover, we might even use an input plan generated by the user as a sequence of activities. It is important to note that this plan may abstract out the scheduling (time+resources) requirements, i.e. the plan does not need to be executable because the objective of the integrated module is precisely to repair a given plan and make it fully executable w.r.t. all the scheduling requirements. To sum up, we can use a pure STRIPS planner or a PDDL planner or a hand-tailored plan. The output will be, in any case, a skeleton plan or causal structure for the problem at hand. Obviously, the more advanced method we use for generating the plan, the better plan quality. • The modelling. The second step in our framework is the problem formulation. Instead of encoding planning structures (a planning graph, for instance) (Kambhampati 2000), we encode the plan causal structure and the additional constraints as a CSP. Thus, this paper shows, in detail, the formulation of plans with complex constraints based on the works of (Refanidis 2005; Vidal & Geffner 2006). We can undertake the modelling of a complete plan as a

Using constraint programming to model complex plans in an integrated approach for planning and scheduling

使用约束规划以集成的规划和调度方法对复杂计划进行建模

文献摘要

被引文献

相似文献