Search and Inference in AI Planning

While Planning has been a key area in Artificial Intelligence since its beginnings, significant changes have occurred in the last decade as a result of new ideas and a more established empirical methodology. In this invited talk, I will focus on Optimal Planning where these new ideas can be understood along two dimensions: branching and pruning. Both heuristic search planners, and SAT and CSP planners can be understood in this way, with the latter branching on variables and pruning by constraint propagation, and the former branching on actions and pruning by lower bound estimations. The two formulations, however, have a lot in common, and some key planners such as Graphplan can be understood in either way: as computing a lower bound function and searching backwards from the goal, or as performing a precise, bounded form of variable elimination, followed by backtracking. The main limitation of older, so-called Partial Ordered Causal Link (POCL) planners, is that they provide smart bra...

• Type: Conference
• Year: 2005
• Where: CP
• Authors: Hector Geffner
• Keywords: CP 2005, Lower Bound Estimations, Lower Bounds, Programming Languages, Pruning