Reformulation Approaches for Numeric Planning

  • Diaeddin Alarnaouti

Student thesis: Doctoral Thesis

Abstract

A characteristic of intelligent behaviour is the ability to reason with knowledge of action and change, in order to synthesize plans to achieve desired goals. Systems utilizing such reasoning abilities can be considered autonomous, as they can choose which actions to perform to achieve previously identified and selected goals. Automated planning (AP) plays a central role in autonomous agent systems, focusing on approaches for producing plans—sequences of actions that need to be performed—that enable an agent to achieve clearly stated goals. To generate plans, a pre-engineered symbolic domain model must be provided to a suitable planning engine. To ensure that AP techniques can generate plans in complex real-world domains, i.e., to support operationality, one of the main strategies that can be exploited in domain-independent planning is ’reformulation.’ Reformulation techniques focus on modifying the way knowledge is encoded in the symbolic model. Many studies have demonstrated the enhancements achieved by reformulating the representation of domain models in classical planning, where the environment is deterministic, fully observable, and involves a single agent. One of the most well-known reformulation approaches is macro-operators. These com-bine primitive operators into one compound operator that can function exactly as the original operators if executed in sequence. Unlike classical planning environments, real-world environments are more complex, which has led to extending the definition language to cover actions’ duration and resource limitations, handled by numeric planning. In this research, the possibility of exploiting the macro-operators reformulation technique in numeric domains will be investigated. We developed the ’Generate Numeric Macro Action Algorithm,’ which aims to create numeric macro-actions following the same pattern as classical macro-operators: planner-independent and domain-independent. The results we obtained shown a promising impact on planning process. Which can make a useful contribution to the planning community and certain applications where planning enhancements is required.
Date of Award18 Dec 2024
Original languageEnglish
SupervisorMauro Vallati (Main Supervisor) & George Bargiannis (Co-Supervisor)

Cite this

'