Pascal Bercher

Résumé:

This talk presents an overview of my recent work on hierarchical task network (HTN) planning, a hierarchical approach to modeling and solving planning problems. After introducing HTN planning and contrasting it with non-hierarchical planning, I introduce one of the current state-of-the-art approaches for solving HTN problems: progression search. A recently identified theoretical drawback of this approach is presented, along with one way it can be resolved. Then, an extension of the typical, deterministic problem setting is presented that operates in Fully Observable Non-Deterministic (FOND) environments, where actions have uncertain effects. Finally, I outline another line of research: model repair -- which deals with the research question of how potentially flawed models can be automatically corrected to ensure desired model and planning behavior.

Bio: 

Pascal Bercher is an ARC DECRA Fellow and Associate Professor at the Australian National University (ANU) in Canberra, Australia. His research focuses on Hierarchical Task Network (HTN) planning -- a subfield of Artificial Intelligence. His main lines of research include theoretical foundations, heuristic search, and model repair. He is an active member of the automated planning community and a founder and the main organizer of HPlan -- the annual ICAPS workshop on Hierarchical Planning.