![]() (Show abstract) (Hide abstract) (PDF) (code, scripts, and data) Detecting Unsolvability Based on Separating Functions. Remo Christen, Salomé Eriksson, Florian Pommerening and Malte Helmert. Our methods outperform it in one benchmark domain. Per-domain learning and the LAMA planner is still dominant but Per-instance learning often yields stronger heuristics than Key lessons are that our methodsĪnd imitation learning are highly complementary that Three different NN heuristic function learning architecturesįor cross-comparison in an experiment of unprecedentedīreadth in this context. We empirically compare these methods to (a) and (b), aligning Search effort instead of goal distance, which as we showĬonverges to the perfect heuristic under idealized circumstances. We introduce a new bootstrapping variant that estimates Here we explore three methodsįor (a) that make training data generation scalable throughīootstrapping and approximate value iteration. Generation, the latter to domains where the necessary knowledge The approach to instances small enough for training data Work addressed this question by (a) per-instance imitation How can we train neural network (NN) heuristic functions forĬlassical planning, using only states as the NN input? Prior (Show abstract) (Hide abstract) (PDF) (code and data) Proceedings of the 32nd International Conference on Automated Planning and Scheduling Neural Network Heuristic Functions for Classical Planning: Bootstrapping and Comparison to Other Methods. ![]() Patrick Ferber, Florian Geißer, Felipe Trevizan, Malte Helmert and Jörg Hoffmann. Raise interesting questions for future research. Heuristics up to a certain problem difficulty. However, inĬontrast to PDBs, Cartesian abstractions yield perfect General Cartesian and merge-and-shrink abstractions. We find that PDBs still outperform the more Newer types of abstraction heuristics compare against patternĭatabase (PDB) heuristics, the state-of-the-art for solving Furthermore, we extend counterexample-guided CartesianĪbstraction refinement (CEGAR) to support factored effect tasks,Ī class of planning tasks with a specific kind of conditionalĮffects which includes Rubik’s Cube. To obtain a concise model, we require conditionalĮffects. Present the first model of Rubik’s Cube for general problem Short sequence of rotations to solve the cube is hard. ![]() Scrambled cube and rotates the six faces until each faceĬontains only stickers of one color. Since its invention in 1974, the Rubik’s Cube puzzle fascinates
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |