Results 1 - 10
of
14
Learning heuristic functions from relaxed plans
- In International Conference on Automated Planning and Scheduling (ICAPS
, 2006
"... We present a novel approach to learning heuristic functions for AI planning domains. Given a state, we view a relaxed plan (RP) found from that state as a relational database, which includes the current state and goal facts, the actions in the RP, and the actions ’ add and delete lists. We represent ..."
Abstract
-
Cited by 22 (7 self)
- Add to MetaCart
(Show Context)
We present a novel approach to learning heuristic functions for AI planning domains. Given a state, we view a relaxed plan (RP) found from that state as a relational database, which includes the current state and goal facts, the actions in the RP, and the actions ’ add and delete lists. We represent heuristic functions as linear combinations of generic features of the database, selecting features and weights using training data from solved problems in the target planning domain. Many recent competitive planners use RP-based heuristics, but focus exclusively on the length of the RP, ignoring other RP features. Since RP construction ignores delete lists, for many domains, RP length dramatically under-estimates the distance to a goal, providing poor guidance. By using features that depend on deleted facts and other RP properties, our learned heuristics can potentially capture patterns that describe where such under-estimation occurs. Experiments in the STRIPS domains of IPC 3 and 4 show that best-first search using the learned heuristic can outperform FF (Hoffmann & Nebel 2001), which provided our training data, and frequently outperforms the top performances in IPC 4.
COMBINING DESCRIPTION LOGIC REASONING WITH AI PLANNING FOR COMPOSITION OF WEB SERVICES
, 2006
"... As Web Services become more prevalent — with the aim of achieving inter-operability between heterogeneous, decentralized and distributed systems — the problem of selecting and composing services to accomplish a given task becomes more important. Using Web ontologies to describe different properties ..."
Abstract
-
Cited by 18 (0 self)
- Add to MetaCart
As Web Services become more prevalent — with the aim of achieving inter-operability between heterogeneous, decentralized and distributed systems — the problem of selecting and composing services to accomplish a given task becomes more important. Using Web ontologies to describe different properties of Web Ser-vices provided by separate developers facilitates their integration. Automating the composition of Web Services is essential for various different subjects ranging from ordinary users performing tasks on the Web, businesses carrying out complex trans-actions, and scientists collaborating with each other on the computational Grid. In this thesis I present the HTN-DL formalism which combines Hierarchical Task Network (HTN) planning and Description Logics (DL) to automatically com-pose Web Services which are described with Web Ontology Language (OWL). The main contributions of this thesis are as follows: • The HTN-DL formalism, which couples Hierarchical Task Network (HTN) planning and Description Logics. HTN-DL combines the expressivity of De-scription Logics with the efficiency of HTN planning systems to solve Web
Learning to do htn planning
- In Proceedings of the Sixteenth International Conference on Automated Planning and Scheduling, 390
, 2006
"... We describe the HDL algorithm, which learns HTN domain representations by examining plan traces produced by an expert problem-solver. Prior work on learning HTN methods required everything to be given in advance except for the methods ’ preconditions, and the learner would learn the preconditions. I ..."
Abstract
-
Cited by 10 (0 self)
- Add to MetaCart
(Show Context)
We describe the HDL algorithm, which learns HTN domain representations by examining plan traces produced by an expert problem-solver. Prior work on learning HTN methods required everything to be given in advance except for the methods ’ preconditions, and the learner would learn the preconditions. In contrast, HDL has no prior information about the methods. In our experiments, in most cases HDL converged fully with no more than about 200 plan traces. Furthermore, even when HDL was only halfway to convergence, it usually was able to produce HTN methods that were sufficient to solve more than 3/4 of the planning problems in the test set.
Transfer Learning of Hierarchical Task-Network Planning Methods in a Real-Time Strategy Game
"... We describe a new integrated and automated AI planning and learning architecture, called Learn2SHOP. Learn2SHOP departs significantly from the previous works on AI planning and learning in that its modular architecture integrates Hierarchical Task Network (HTN) planning, concept learning, and comput ..."
Abstract
-
Cited by 3 (1 self)
- Add to MetaCart
(Show Context)
We describe a new integrated and automated AI planning and learning architecture, called Learn2SHOP. Learn2SHOP departs significantly from the previous works on AI planning and learning in that its modular architecture integrates Hierarchical Task Network (HTN) planning, concept learning, and computer simulations. Using simulations during the planning and learning process enables the system to get information about the outcomes of the actions. We have implemented Learn2SHOP and tested it on a transfer-learning task. The objective of transfer learning is transferring knowledge and skills learned from a wide variety of previous situations to the current, and likely different, previously unencountered problems(s). The experiments with Learn2SHOP have demonstrated the advantages of integrating planning, learning, and simulation in a real-time strategy game engine.
Learning Recursive HTN-Method Structures for planning
"... HTN planning is one of the most effective planning methods in AI. However, designing the HTN-decomposition methods is a very difficult task which has been achieved mainly by humans. It would therefore be desirable to design automated learning methods to acquire these decomposition methods from obser ..."
Abstract
-
Cited by 3 (0 self)
- Add to MetaCart
HTN planning is one of the most effective planning methods in AI. However, designing the HTN-decomposition methods is a very difficult task which has been achieved mainly by humans. It would therefore be desirable to design automated learning methods to acquire these decomposition methods from observed action sequences. In this work, we explore how to apply model-based clustering in order to construct task decomposition hierarchies and summarize a database of action sequences. We present a probabilistic model for unsupervised learning of HTN methods from action sequences. Based on this model, we introduce a novel two-pronged approach by simultaneously learning a Markov model for action segment clusters from action sequences and then learning an action parameter model for recognizing tasks. These models are integrated together to construct action clusters. Then, an abstraction algorithm is applied to extract variables from the action parameters in each cluster to obtain succinct HTN methods. We introduce evaluation metrics for this approach, and test the algorithm in a logistics planning domain.
Opmaker2: Efficient action schema acquisition
- In Proceedings of the 26th Workshop of the UK Planning and Scheduling Special Interest Group
, 2007
"... The problem of formulating knowledge bases containing specifications of dynamic knowledge is a barrier to the widespread uptake of AI planning. Machine learning has been used with some success in the past, but the inputs re-quired are either too detailed, or the learning process has required many ex ..."
Abstract
-
Cited by 3 (0 self)
- Add to MetaCart
(Show Context)
The problem of formulating knowledge bases containing specifications of dynamic knowledge is a barrier to the widespread uptake of AI planning. Machine learning has been used with some success in the past, but the inputs re-quired are either too detailed, or the learning process has required many examples. Further, learning has been confined to propositional ac-tions or parts of actions such as preconditions. The field of ontological engineering has had an impact on the wider community in that appli-cation ontologies (which contain ”static ” struc-tural knowledge of applications) are becoming widespread. Here we introduce a methodology that is based on the existence of a strong struc-tural model of an application. Using a small number of user training sequences, we illus-trate how the method can induce action schema and compound methods. To do this we extend GIPO’s Opmaker system so that it can induce actions from training sequences without inter-mediate state information and without requir-ing large numbers of examples. This method shows the potential for considerably reducing the burden of knowledge engineering, in that it would be possible to embed the method into an autonomous program (agent) which required to do planning. We illustrate the algorithm as part of an overall method to induce structured domain model, and comment on initial results that show the efficacy of the induced model em-pirically.
Learning Plan Knowledge in Tasks with Information Gathering
- In Acquiring Planning Knowledge via Demonstration, Papers from the 2007 AAAI Workshop (Tech. Report WS-07-02), 26–31. Menlo Park
"... In this paper we describe a framework for learning plan knowledge using expert solution traces in domains that include information-gathering tasks. We describe an extension to a special class of hierarchical task networks (HTNs) that can naturally represent information-gathering tasks and partial pl ..."
Abstract
-
Cited by 1 (1 self)
- Add to MetaCart
In this paper we describe a framework for learning plan knowledge using expert solution traces in domains that include information-gathering tasks. We describe an extension to a special class of hierarchical task networks (HTNs) that can naturally represent information-gathering tasks and partial plans. We show how an existing analytical learning algorithm designed to learn a special form of HTNs can be improved to address the issues raised in information gathering. We also describe how our learning algorithm can use facts the expert explicitly asserts during task execution. Finally we report the preliminary evaluation of our system on a web form based scheduling and information-gathering domain.
Learning Game Strategies by Experimentation
"... Deliberative experimental learning is an approach for learning explicit game strategies in a small number of trials by posting and experimentally satisfying learning goals. Learning explicit strategies is important for producing knowledge that can easily be transferred via analogy to new games, as w ..."
Abstract
-
Cited by 1 (1 self)
- Add to MetaCart
Deliberative experimental learning is an approach for learning explicit game strategies in a small number of trials by posting and experimentally satisfying learning goals. Learning explicit strategies is important for producing knowledge that can easily be transferred via analogy to new games, as well as for rapid learning. In our approach, experiments, or plans for learning, serve to drive both exploration and credit assignment, by helping to explain the execution trace. We describe a system that learns strategic plans for a subset of games in the General Game Playing (GGP) framework and present experimental results showing that it learns to win most of these games in fewer than 10 trials. 1
Distribution: Public Status: Final
"... This document addresses Knowledge Representation issues for the development of COMPANIONS, which are seen as Embodied Conversational Agents possessing cognitive abilities that make them believable, as well as user-friendly, in their role of assistants to daily activities. We first review the knowled ..."
Abstract
- Add to MetaCart
This document addresses Knowledge Representation issues for the development of COMPANIONS, which are seen as Embodied Conversational Agents possessing cognitive abilities that make them believable, as well as user-friendly, in their role of assistants to daily activities. We first review the knowledge models implemented in assistive cognitive systems and draw conclusions on the possibility to unify the representation of user activities and domain knowledge used Planning technologies. We then introduce our choice of Planning formalism and how it could support multiple modes of interventions for a COMPANION. Representing user tasks and activities may not suffice to capture the complexity of certain domains, which require additional modelling in the traditional knowledge acquisition sense, so as to support domain inference in those areas where COMPANIONS are meant to assist (such as elementary “Medical ” knowledge in the area of Healthy lifestyles). Finally, because COMPANIONS are meant to be embodied and situated in the user’s environment, it is necessary to discuss how their knowledge representations can be
