No abstract

Extending the classical planning formalism with state-dependent action costs (SDAC) allows an up to exponentially more compact task encoding. Recent work proposed to use edge-valued multi-valued decision diagrams (EVMDDs) to represent cost functions, which allows to automatically detect and exhibit structure in cost functions and to make heuristic estimators accurately reflect SDAC. However, so far only the inadmissible additive heuristic has been considered in this context. In this paper, we define informative admissible abstraction heuristics which enable optimal planning with SDAC. We discuss how abstract cost values can be extracted from EVMDDs that represent concrete cost functions without adjusting them to the selected abstraction. Our theoretical analysis shows that this is efficiently possible for abstractions that are Cartesian or coarser. We adapt the counterexample-guided abstraction refinement approach to derive such abstractions. An empirical evaluation of the resulting heuristic shows that highly accurate values can be computed quickly.

How can we train neural network (NN) heuristic functions for classical planning, using only states as the NN input? Prior work addressed this question by (a) per-instance imitation learning and/or (b) per-domain learning. The former limits the approach to instances small enough for training data generation, the latter to domains where the necessary knowledge generalizes across instances. Here we explore three methods for (a) that make training data generation scalable through bootstrapping and approximate value iteration. In particular, we introduce a new bootstrapping variant that estimates search effort instead of goal distance, which as we show converges to the perfect heuristic under idealized circumstances. We empirically compare these methods to (a) and (b), aligning three different NN heuristic function learning architectures for cross-comparison in an experiment of unprecedented breadth in this context. Key lessons are that our methods and imitation learning are highly complementary; that per-instance learning often yields stronger heuristics than per-domain learning; and the LAMA planner is still dominant but our methods outperform it in one benchmark domain.

Symbolic search has proven to be a competitive approach to cost-optimal planning, as it compactly represents sets of states by symbolic data structures. While heuristics for symbolic search exist, symbolic bidirectional blind search empirically outperforms its heuristic counterpart and is therefore the dominant search strategy. This prompts the question of why heuristics do not seem to pay off in symbolic search. As a first step in answering this question, we investigate the search behaviour of symbolic heuristic search by means of BDDA⋆. Previous work identified the partitioning of state sets according to their heuristic values as the main bottleneck. We theoretically and empirically evaluate the search behaviour of BDDA⋆ and reveal another fundamental problem: we prove that the use of a heuristic does not always improve the search performance of BDDA⋆. In general, even the perfect heuristic can exponentially deteriorate search performance.

Symbolic representations have attracted significant attention in optimal planning. Binary Decision Diagrams (BDDs) form the basis for symbolic search algorithms. Closely related are Algebraic Decision Diagrams (ADDs), used to represent heuristic functions. Also, progress was made in dealing with models that take state-dependent action costs into account. Here, costs are represented as Edge-valued Multi-valued Decision Diagrams (EVMDDs), which can be exponentially more compact than the corresponding ADD representation. However, they were not yet considered for symbolic planning. In this work, we study EVMDD-based symbolic search for optimal planning. We define EVMDD-based representations of state sets and transition relations, and show how to compute the necessary operations required for EVMDD-A*. This EVMDD-based version of symbolic A* generalizes its BDD variant, and allows to solve planning tasks with state-dependent action costs. We prove theoretically that our approach is sound, complete and optimal. Additionally, we present an empirical analysis comparing EVMDD-A* to BDD-A* and explicit A* search. Our results underscore the usefulness of symbolic approaches and the feasibility of dealing with models that go beyond unit costs.