Adaptive Subgoal Search (AdaSubS) addresses the fact that states within one problem can require different amounts of look-ahead. It generates candidate subgoals at several distances and uses reachability verification to filter infeasible states, moving between longer and shorter planning horizons as needed.
The method was evaluated on Sokoban, the Rubik’s Cube, and the INT inequality-proving benchmark. This work formed the central research contribution of my MSc and established the adaptive-planning direction that I continue to pursue.