Learning, Representing and Exploiting Sequential Nogoods in Planning


In Artificial Intelligence, when you are thinking-about-what-you-are-going-to-do, we say you are "planning". This project is about automating planning. If you are successful in this project, then people will not have to think so hard, as they will have computer systems to do it for them. Such amaze, much wow!

Automated planning systems work by constructing plans, bit by bit. During that process it is not uncommon for a planning system to determine that something it has been thinking about is wrong. Automated planning systems are notorious for "thinking about" the same completelly wrong thing over and over again.  This project is about synthesizing concise and general explanations about what went wrong. The explanations will be used to minimise the amount of time the planning system spends thinking about wrong stuff. The focus will be on learning compact representations of action sequences that are wrong.


 - Funding - 
Australian citizens who undertake this project may be eligible for a generous stipend. Please contact us to find out more.



learn nogood sequences and represent them compactly in such a way that they can be efficiently exploited by a plansearch procedure.


passion for artificial intelligence, programming, and interest in research.

Background Literature

* M. Krajnanský, J. Hoffmann, O. Buffet and A. Fern. Learning pruning rules for heuristic search planning. ECAI 2014.

* M. Rankooh and G. Ghassem-Sani. ITSAT: An efficient SAT-based temporal planner. JAIR, 2015.

* M. Steinmetz and J. Hoffmann, State Space Search Nogood Learning:  Online Refinement of Critical-Path Dead-End Detectors in Planning. AIJ, 2017.

* Toby O. Davies, Adrian R. Pearce, Peter J. Stuckey, Nir Lipovetzky. Sequencing Operator Counts. ICAPS, 2015.




research experience, and knowledge of advanced AI topics.


artificial intelligence, automated planning, nogoods

Updated:  10 August 2021/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing