Learning to solve problems by searching for macro-operators
Learning to solve problems by searching for macro-operators
Learning by analogical reasoning in general problem-solving
Learning by analogical reasoning in general problem-solving
Artificial intelligence: a new synthesis
Artificial intelligence: a new synthesis
On reasonable and forced goal orderings and their use in an agenda-driven planning algorithm
Journal of Artificial Intelligence Research
Hybrid STAN: Identifying and managing combinatorial optimisation sub-problems in planning
IJCAI'01 Proceedings of the 17th international joint conference on Artificial intelligence - Volume 1
Local search topology in planning benchmarks: an empirical analysis
IJCAI'01 Proceedings of the 17th international joint conference on Artificial intelligence - Volume 1
Learning Control Knowledge for Forward Search Planning
The Journal of Machine Learning Research
Learning measures of progress for planning domains
AAAI'05 Proceedings of the 20th national conference on Artificial intelligence - Volume 3
| Hi-index | 0.00 |
Heuristic search planners are so far the most successful. Almost all use as their heuristic an estimate of the distance to a goal state. We formalize a logical measure of progress, defined as a predicate P(x,s) true of objects x at a situation s. Actions which increase P's extension are guaranteed to move closer to a goal situation, so that P enables us to form plans without search. One example of a measure of progress is the concept of final position Used in BlocksWorld. It is not clear how to find a P for an arbitrary domain, so instead we identify three different classes of domains and conditions which allow us to construct a measure of progress.An obvious P will not deliver optimal plans, but it should encode plans which are "good enough" Our paradigm is entirely within first-order logic, allowing us to extend our results to concurrent domains and those containing non-trivial state constraints. It turns out P not only encodes goal orderings, but subgoal orderings. P also gives rise to a strategy function a(s) which can be used to create a universal (complete) teleo-reactive (TR) program. Given the fact that P-increasing actions will never require backtracking, this TR program can be a powerful on-line planner.