TITLE: Fast Planning with Macro-Operators

PRESENTER: Adi Botea, http://www.cs.ualberta.ca/~adib, adib@cs.ualberta.ca

AFFILIATION:Department of Computing Science, University of Alberta, Canada, http://www.cs.ualberta.ca/

DATE: Friday 7th October 2005

TIME: 12:45:00

PLACE: CSE Seminar Room, Level 1, K17

ABSTRACT:

Application-specific knowledge often has a great contribution to the
success of AI programs. However, in domain-independent planning,
such knowledge is not available beforehand, since one planner
addresses many domains, including new ones.

This talk presents two methods that infer new information
about a domain, and encode it in the form of macro-operators.
Implemented on top of the state-of-the-art planner FF, this strategy can
lead to an improvement by orders of magnitude, for domains where
structural information can automatically be acquired.
The resulting planner, Macro-FF, successfully participated in the
International Planning Competition IPC-4, taking the
first place in 3 out of 7 domains where it competed.

BIOGRAPHY OF SPEAKER:

Adi Botea is expected to obtain a PhD degree from the University of
Alberta, Canada by the end of 2005. He has completed a MSc program in
Computer Science at the University of Bucharest, Romania, with a GPA of
98%, the best in the class of 1998. Adi has worked in industry as a software
engineer between 1999 and 2000. His research interests include AI
planning, heuristic search, and AI in games.

Host:

Alfredo Gabaldon

Seminar Convenor:

Van Hai Ho