Planning with Derived Predicates Through Rule-Action Graphs and Local Search Techniques
暂无分享,去创建一个
In classical domain-independent planning, derived predicates are predicates that the domain actions can only indirectly affect. Their truth in a state can be inferred by particular axioms, that enrich the typical operator description of a planning domain.
As discussed in [3,6], derived predicates are practically useful to express in a concise and natural way some indirect action effects, such as updates on the transitive closure of a relation. Moreover, compiling them away by introducing artificial actions and facts in the formalization is infeasible because, in the worst case, we have an exponential blow up of either the problem description or the plan length [6]. This suggests that is worth investigating new planning methods supporting derived predicates, rather than using existing methods with “compiled” problems.
[1] Bernhard Nebel,et al. In Defense of PDDL Axioms , 2003, IJCAI.
[2] Ivan Serina,et al. Planning Through Stochastic Local Search and Temporal Action Graphs in LPG , 2003, J. Artif. Intell. Res..
[3] A. Gerevini,et al. Planning with Derived Predicates through Rule-Action Graphs and Relaxed-Plan Heuristics , 2005 .