Recent discoveries in automated planning are broadening the scope of planners, from toy problems to real applications. However, applying automated planners to real-world problems is far from simple. On the one hand, the definition of accurate action models for planning is still a bottleneck. On the other hand, off-the-shelf planners fail to scale-up and to provide good solutions in many domains. In these problematic domains, planners can exploit domain-specific control knowledge to improve their performance in terms of both speed and quality of the solutions. However, manual definition of control knowledge is quite difficult. This paper reviews recent techniques in machine learning for the automatic definition of planning knowledge. It has been organized according to the target of the learning process: automatic definition of planning action models and automatic definition of planning control knowledge. In addition, the paper reviews the advances in the related field of reinforcement learning.

/pdf/review-a-review-of-machine-learning-for-automated-planning-170ctue9g9.pdf

Review: a review of machine learning for automated planning

There are many approaches for solving planning problems. Many of these approaches are based on ‘brute force’ search methods and they usually do not care about structures of plans previously computed in particular planning domains. By analyzing these structures, we can obtain useful knowledge that can help us find solutions to more complex planning problems. The method described in this paper is designed for gathering macro-operators by analyzing training plans. This sort of analysis is based on the investigation of action dependencies in training plans. Knowledge gained by our method can be passed directly to planning algorithms to improve their efficiency

/pdf/generation-of-macro-operators-via-investigation-of-action-6dcqak6ppt.pdf

Generation of macro-operators via investigation of action dependencies in plans

When AI search methods are applied to software process models, then appropriate technologies can be discovered for a software project. We show that those recommendations are greatly affected by the business context of its use. For example, the automatic defect reduction tools explored by the ASE community are only relevant to a subset of software projects, and only according to certain value criteria. Therefore, when arguing for the value of a particular technology, that argument should include a description of the value function of the target user community.

/pdf/understanding-the-value-of-software-engineering-technologies-12ob07novy.pdf

Understanding the Value of Software Engineering Technologies

At the beginning of the new century, Information Technologies had become basic and indispensable
constituents of the production and preparation processes for all kinds of goods and services and
with that are largely influencing both the working and private life of nearly every citizen. This
development will continue and even further grow with the continually increasing use of the Internet
in production, business, science, education, and everyday societal and private undertaking.
Recent years have shown, however, that a dramatic enhancement of software capabilities is required,
when aiming to continuously provide advanced and competitive products and services in all these
fast developing sectors. It includes the development of intelligent systems – systems that are more
autonomous, flexible, and robust than today’s conventional software.
Intelligent Planning and Scheduling is a key enabling technology for intelligent systems. It has
been developed and matured over the last three decades and has successfully been employed for a
variety of applications in commerce, industry, education, medicine, public transport, defense, and
government.
This document reviews the state-of-the-art in key application and technical areas of Intelligent Planning
and Scheduling. It identifies the most important research, development, and technology transfer
efforts required in the coming 3 to 10 years and shows the way forward to meet these challenges in
the short-, medium- and longer-term future.
The roadmap has been developed under the regime of PLANET – the European Network of Excellence
in AI Planning. This network, established by the European Commission in 1998, is the co-ordinating
framework for research, development, and technology transfer in the field of Intelligent Planning and
Scheduling in Europe.
A large number of people have contributed to this document including the members of PLANET non-
European international experts, and a number of independent expert peer reviewers. All of them are
acknowledged in a separate section of this document.
Intelligent Planning and Scheduling is a far-reaching technology. Accepting the challenges and progressing
along the directions pointed out in this roadmap will enable a new generation of intelligent
application systems in a wide variety of industrial, commercial, public, and private sectors.

Technological roadmap on AI planning and scheduling

Domain independent planning engines accept a planning task description in a language such as PDDL and return a solution plan. Performance of planning engines can be improved by gathering additional...

/pdf/outer-entanglements-a-general-heuristic-technique-for-3late026fg.pdf

Outer entanglements: a general heuristic technique for improving the efficiency of planning algorithms

This paper is concerned with state space problem solvers that achieve generality by learning strong heuristics through experience in a particular domain. We specifically consider two ways of learning by analysing past solutions that can improve future problem solving: creating macros and the chunks. A method of learning search heuristics is specified which is related to 'chunking' but which complements the use of macros within a goal directed system. An example of the creation and combined use of macros and chunks, taken from an implemented system, is described.

/pdf/combining-weak-learning-heuristics-in-general-problem-4md37piild.pdf

T. L. McCluskey

Papers

Combining weak learning heuristics in general problem solvers