The paper "Decoupled Search for the Masses: A Novel Task Transformation for Classical Planning" has won the Best Paper Award at the 2024 International Conference on Automated Planning and Scheduling (ICAPS). The paper is authored by David Speck and Daniel Gnad of the Machine Reasoning Lab, Division Artificial Intelligence and Integrated Computer Systems at the Department of Computer and Information Science.
Paper description by the authors
Automated planning is an important topic in artificial intelligence and a core research area of the Machine Reasoning Lab at Linköping University, Sweden, led by Jendrik Seipp.
Planning itself can be described as the art and practice of thinking before acting. Modern artificial intelligence systems typically try to find a plan for a given planning task by searching through the possible states of the world until a sequence of actions is found that transforms the current state into a goal state. Prominent examples are satellite control or logistics problems. Since such problems typically involve more possible states than atoms in the universe, several paradigms exist to solve these planning tasks more efficiently.
One such paradigm is called decoupled search, which attempts to automatically decouple a given planning problem based on causal relationships. This means that in a satellite control problem, one can automatically deduce that different components of the satellite can be controlled independently, while synchronizing these components in a certain way. However, this approach has the disadvantage that it requires specialized search algorithms, and recent advances in automated planning cannot be directly applied to this type of automated reasoning. In our work, we have shown theoretically and empirically that one can simulate decoupled search by transforming the planning problem description. In this way, no special algorithms are required to causally decompose a planning task into separate components, and the full existing and future toolbox of planning research can be readily used together with this approach.
Read more:
- The paper and supplementary material can be downloaded here.
- The ICAPS Conference 2024 - a forum for exchanging news and research results on the theory and applications of intelligent and automated planning and scheduling technology.