On Managing Temporal Information for Handling Durative Actions in LPG

LPG is a recent planner based on local search and planning graphs which supports durative actions specified by the new standard language pddl2.1. The planner was awarded at the 3rd Planning Competition (Toulouse, 2002) for its very good performance. This paper focuses on how lpg represents and manages temporal information to handle durative actions during the construction of a plan. In particular, we introduce a plan representation called Temporal Durative Action Graph (TDA-graph) which distinguishes different types of constraints for ordering the actions and allows to generate plans with a high degree of parallelism. An experimental analysis shows that the techniques here presented are effective, and that in temporal domains our planner outperforms all other fully-automated planners that took part in the contest.