Enhancing SAT Based Planning with Landmark Knowledge

Enhancing SAT Based Planning with Landmark Knowledge

Elffers, J.
Konijnenberg, D.
Walraven, E.M.P.
Spaan, M.T.J.

Electrical Engineering, Mathematics and Computer Science

Software Computer Technology

2013-11-07

Several approaches exist to solve Artificial Intelligence planning problems, but little attention has been given to the combination of using landmark knowledge and satisfiability (SAT). Landmark knowledge has been exploited successfully in the heuristics of classical planning. Recently it was also shown that landmark knowledge can improve the performance of SAT based planners, but it was unclear how and in which domains they were effective. We investigate the relationship between landmarks and plan generation performance in SAT. We discuss a recently proposed heuristic for planning using SAT and suggest improvements. We compare the effects of landmark knowledge in parallel and sequential planning, also looking at previous research. It turns out that landmark knowledge can be beneficial, but performance highly depends on the planning domain and the planning problem itself.

http://resolver.tudelft.nl/uuid:a936537f-8090-46d3-8b53-59f3ee1dbdc2

BNAIC 2013: 25th Belgium-Netherlands Conference on Artificial Intelligence, Delft, The Netherlands, 7-8 November 2013

Institutional Repository

conference paper

(c) 2013 The Author(s)