The Real-time Heuristic Search project focuses on heuristic search
methods where the planning time per each action is upper-bounded
independently of the problem size. Applications include video game
pathfinding and moving target pursuit. We have been studying real-time
heuristic since 2003 and made several contributions including a
discovery of lookahead pathology and ways to address it by selecting
the amount of planning time dynamically per action. We have also
designed real-time heuristic search algorithms that use state
abstraction to speed up learning and search. An unorthodox application
of our search methods was to discover compact and powerful sets of
actions for an MDP solver. Our current interests include coordinated
algorithms for real-time multi-unit moving target pursuit, a
theoretical analysis of real-time heuristic search methods including
those for automated action set selection, new methods for dynamic
control and novel applications such as to emotion modeling and
cellular automata evolution. Our partners on this project are the
GAMES group and Russ Greiner at the University of Alberta, Yngvi
Björnsson and his group at Reykjavik University, Mitja Luštrek at
Jožef Stefan Institute,  Sven Koenig at the University of Southern
California and Bioware Corp.