Landmarks, Critical Paths and Abstractions: What's the Difference Anyway?

Self Cite

Potential heuristics, recently introduced by Pommerening et al., characterize admissible and consistent heuristics for classical planning as a set of declarative constraints. Every feasible solution for these constraints defines an admissible heuristic, and we can obtain heuristics that optimize certain criteria such as informativeness by specifying suitable objective functions. The original paper only considered one such objective function: maximizing the heuristic value of the initial state. In this paper, we explore objectives that attempt to maximize heuristic estimates for all states (reachable and unreachable), maximize heuristic estimates for a sample of reachable states, maximize the number of detected dead ends, or minimize search effort. We also search for multiple heuristics with complementary strengths that can be combined to obtain even better heuristics.

Section: Automatic Diversificationmentioning

confidence: 99%

New Optimization Functions for Potential Heuristics

Seipp

Pommerening

Helmert

2015

Self Cite

“…This works considers three baseline domain-independent heuristics which are based on the delete-relaxation: h max (admissible), h add (inadmissible) (Bonet and Geffner 2001), and the Landmark-Cut heuristic (admissible) (Helmert and Domshlak 2009).…”

Section: Planning Heuristicsmentioning

confidence: 99%

“…Geffner 2001; Helmert and Domshlak 2009), of which h max , h add , and LM-cut are popular examples. These heuristics can be seen as the least-cost path in the hypergraph representing the delete-relaxed problem for a suitable aggregation function.…”

Section: Introductionmentioning

confidence: 99%

Learning Domain-Independent Planning Heuristics with Hypergraph Networks

Shen

Trevizan

Thiébaux

2020

We present the first approach capable of learning domain-independent planning heuristics entirely from scratch. The heuristics we learn map the hypergraph representation of the delete-relaxation of the planning problem at hand, to a cost estimate that approximates that of the least-cost path from the current state to the goal through the hypergraph. We generalise Graph Networks to obtain a new framework for learning over hypergraphs, which we specialise to learn planning heuristics by training over state/value pairs obtained from optimal cost plans. Our experiments show that the resulting architecture, STRIPS-HGNs, is capable of learning heuristics that are competitive with existing delete-relaxation heuristics including LM-cut. We show that the heuristics we learn are able to generalise across different problems and domains, including to domains that were not seen during training.

“…During execution, we kept track of intermediate results, allowing us to examine the maximal reduction achieved by each budget level. We used the Fast Downward planning system (Helmert 2006) running A * with the LM-CUT heuristic (Helmert and Domshlak 2009) for all but the ISS domain for which the IPDB heuristic (Haslum et al 2007) was used. Experiments were run on Intel(R) Xeon(R) CPU X5690 machines, with a time limit of 30 minutes and memory limit of 2 GB.…”

Section: Empirical Evaluationmentioning

confidence: 99%

Strong Stubborn Sets for Efficient Goal Recognition Design

Keren

Gal

Karpas

2018

Goal Recognition Design (GRD) is the task of redesigning environments (either physical or virtual) to allow efficient online goal recognition. In this work we formulate the redesign problem as an optimization problem, aiming at early goal recognition. To this end, we use a measure of worst case distinctiveness (wcd), which represents the maximal number of steps an agent may take before his goal is revealed. With the objective ofminimizing wcd, we construct a search space in which each node in the space is a goal recognition model (one of which is the original model given as input) and one can move from one model to another by applying a model modification, chosen from a set of allowed modifications given as input. Our specific contribution in this work includes the specification of a class of modifications for which we can prune the search space using strong stubborn sets. Such positioning allows reducing the computational overhead of design while preserving completeness. We show that the proposed modification class generalizes previous works in goal recognition design and enriches the state-of-the-art with new modifications for which strong stubborn set pruning is safe. We support our approach by an empirical evaluation that reveals the performance gain brought by the proposed pruning strategy in different goal recognition design settings.