Macro-FF: Improving AI Planning with Automatically Learned Macro-Operators

Botea, Adi; Enzenberger, Markus; Müller, Martin; Schaeffer, Jonathan

doi:10.1613/jair.1696

Cited by 120 publications

(147 citation statements)

References 24 publications

Supporting

Mentioning

145

Contrasting

Order By: Relevance

“…A specific experimental analysis is also needed for having a better understanding of the impact of problems reformulation on the different planning systems; a system for predicting this impact would lead to a great reduction of the learning time needed for selecting the algorithm to use for a specific domain. Moreover, we are interested in combining the approach used for reformulating planning problems with existing techniques for generating macrooperators (e.g., Wizard [16], Macro-FF [1]). …”

Section: Discussionmentioning

confidence: 99%

“…Hence, it is reasonable to assemble these operators into a macro-operator unstack-putdown(?x ?y). Creating macro-operators, which can be understood as 'short-cuts' in the state space, is therefore a well known and studied approach which in some cases can speed up plan generation considerably [1], [16]. Macro-operators can be added into planning domains and reformulated domains can be passed to any planning engine.…”

Section: A Macro-operatorsmentioning

confidence: 99%

See 1 more Smart Citation

An Automatic Algorithm Selection Approach for Planning

Vallati

Chrpa

Kitchin

2013

2013 IEEE 25th International Conference on Tools With Artificial Intelligence

View full text Add to dashboard Cite

Abstract-Despite the advances made in the last decade in automated planning, no planner outperforms all the others in every known benchmark domain. This observation motivates the idea of selecting different planning algorithms for different domains. Moreover, the planners' performances are affected by the structure of the search space, which depends on the encoding of the considered domain. In many domains, the performance of a planner can be improved by exploiting additional knowledge, extracted in the form of macro-operators or entanglements.In this paper we propose ASAP, an automatic Algorithm Selection Approach for Planning that: (i) for a given domain initially learns additional knowledge, in the form of macrooperators and entanglements, which is used for creating different encodings of the given planning domain and problems, and (ii) explores the 2 dimensional space of available algorithms, defined as encodings-planners couples, and then (iii) selects the most promising algorithm for optimising either the runtimes or the quality of the solution plans.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: A Macro-operatorsmentioning

confidence: 99%

An Automatic Algorithm Selection Approach for Planning

Vallati

Chrpa

Kitchin

2013

2013 IEEE 25th International Conference on Tools With Artificial Intelligence

View full text Add to dashboard Cite

show abstract

“…Clement et al assume that such conflicts can be resolved during the scheduling phase, by inserting an available (concurrent) plan-possibly one belonging to a different agent-that asserts a suitable post-condition. 4 We disallow such conflicts, and define a "local" notion of a contingent condition which does not rely on other concurrent plans.…”

Section: Related Workmentioning

confidence: 99%

Reasoning About the Executability of Goal-Plan Trees

Yao

Silva

Logan

2016

Engineering Multi-Agent Systems

View full text Add to dashboard Cite

Abstract. User supplied domain control knowledge in the form of hierarchically structured agent plans is at the heart of a number of approaches to reasoning about action. This knowledge encodes the "standard operating procedures" of an agent for responding to environmental changes, thereby enabling fast and effective action selection. This paper develops mechanisms for reasoning about a set of hierarchical plans and goals, by deriving "summary information" from the conditions on the execution of the basic actions forming the "leaves" of the hierarchy. We provide definitions of necessary and contingent pre-, in-, and postconditions of goals and plans that are consistent with the conditions of the actions forming a plan. Our definitions extend previous work with an account of both deterministic and non-deterministic actions, and with support for specifying that actions and goals within a (single) plan can execute concurrently. Based on our new definitions, we also specify requirements that are useful in scheduling the execution of steps in a set of goal-plan trees. These requirements essentially define conditions that must be protected by any scheduler that interleaves the execution of steps from different goal-plan trees.

show abstract

“…• Learning Macro-actions (Botea, Enzenberger, Müller, & Schaeffer, 2005;Coles & Smith, 2007) are the combination of two or more operators that are considered as new domain operators in order to reduce the search tree depth. However, this benefit decreases with the number of new macro-actions added because they enlarge the branching factor of the search tree causing the utility problem (Minton, 1990).…”

Section: Related Workmentioning

confidence: 99%

Scaling up Heuristic Planning with Relational Decision Trees

Rosa¹,

Jiménez²,

Fuentetaja³

et al. 2011

jair

View full text Add to dashboard Cite

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.

show abstract

Macro-FF: Improving AI Planning with Automatically Learned Macro-Operators

Cited by 120 publications

References 24 publications

An Automatic Algorithm Selection Approach for Planning

An Automatic Algorithm Selection Approach for Planning

Reasoning About the Executability of Goal-Plan Trees

Scaling up Heuristic Planning with Relational Decision Trees

Contact Info

Product

Resources

About