Scaling up Heuristic Planning with Relational Decision Trees

Abstract

Current evaluation functions for heuristic planning are expensive to compute. In numerous planning problems these functions provide good guidance to the solution, so they are worth the expense. However, when evaluation functions are misguiding or when planning problems are large enough, lots of node evaluations must be computed, which severely limits the scalability of heuristic planners. In this paper, we present a novel solution for reducing node evaluations in heuristic planning based on machine learning. Particularly, we define the task of learning search control for heuristic planning as a relational classification task, and we use an off-the-shelf relational classification tool to address this learning task. Our relational classification task captures the preferred action to select in the different planning contexts of a specific planning domain. These planning contexts are defined by the set of helpful actions of the current state, the goals remaining to be achieved, and the static predicates of the planning task. This paper shows two methods for guiding the search of a heuristic planner with the learned classifiers. The first one consists of using the resulting classifier as an action policy. The second one consists of applying the classifier to generate lookahead states within a Best First Search algorithm. Experiments over a variety of domains reveal that our heuristic planner using the learned classifiers solves larger problems than state-of-the-art planners.

A case-based approach to heuristic planning

“…Accordingly, typed sequences can be complemented with other techniques that handle node evaluation issues. As future work we are planning to use CBR node recommendations to validate the action policy of ROLLER [9]. The new algorithm will propose a set of candidate actions to be applied using the relational decision trees, and then will select the actions that replay a typed sequence stored in the case base.…”

“…It performed poorly in the competition because the algorithm (not presented in this work) strongly depends on the quality of the relaxed plan. Other systems such as OBTUSEWEDGE [27], and ROLLER [9], performed better. These systems learn a generalized policy [19], which is a function that maps a metastate to the action that should be applied.…”

Robots in Retirement Homes: Applying Off-the-Shelf Planning and Scheduling to a Team of Assistive Robots

“…For example, TLPlan (Bacchus & Kabanza, 2000) and TALPlan (Kvarnström & Doherty, 2000) make use of control rules to prune the search space while hierarchical task network planners such as O-Plan (Tate, Drabble, & Kirby, 1994) and SHOP2 (Nau et al, 2003) use domain-specific knowledge for decomposing tasks into subtasks. It is also worth mentioning the work done by the planning and learning community who are looking to learn search control knowledge for planning automatically (de la Rosa, Jiménez, Fuentetaja, & Borrajo, 2011;Krajanský, Hoffmann, Buffet, & Fern, 2014) (see also the learning track of the international planning competition). The justification for the restriction of search control knowledge is that human planners are able to solve problems without being given such domain-specific rules.…”

Inner entanglements: Narrowing the search in classical planning by problem reformulation

“…6 Symmetry breaking reuses information about one object to its symmetric "twin" in such a way that "bad" states of one object can be avoided for its symmetric "twin." 10 With growing interest in extracting DCK automatically, emphasis was given on exploiting machine learning techniques that can acquire useful DCK, usually, by analyzing "training plans," which are solutions of simple planning tasks. 4 Another way how performance of planning engines can be improved is by gathering domain control knowledge (DCK), ie, additional knowledge about planning tasks indicating how solution plans would look like.…”

“…DCK can be expressed, for instance, in the form of control rules, 8 temporal logic formulas, 9 or decision trees. 10 With growing interest in extracting DCK automatically, emphasis was given on exploiting machine learning techniques that can acquire useful DCK, usually, by analyzing "training plans," which are solutions of simple planning tasks. This motivated the foundation of the learning track in the IPC, which has been organized since 2008.…”