Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Subgoaling Techniques for Satisficing and Optimal Numeric Planning
Authors: Enrico Scala, Patrik Haslum, Sylvie Thiébaux, Miquel Ramirez
JAIR 2020 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experimental results validate the theoretical assumptions, and indicate that subgoaling substantially improves on the state of the art in optimal and satisficing numeric planning via forward state-space search. |
| Researcher Affiliation | Academia | Enrico Scala EMAIL Universit a di Brescia (Italy) The Australian National University (Australia) Patrik Haslum EMAIL Sylvie Thi ebaux EMAIL The Australian National University (Australia) Miquel Ramirez EMAIL University of Melbourne (Australia) |
| Pseudocode | Yes | Alg. 1 shows a polynomial algorithm proving the previous. The procedure is basically a blind search over the relaxed reachable actions of the problem. Algorithm 1: ˆhmax/add hbd computation... Algorithm 2: Computing hgen(s, G) |
| Open Source Code | Yes | The implementation is part of the ENHSP planning system (https://sites.google.com/view/enhsp/), which is a JAVA implementation containing all the necessary functionalities developed. |
| Open Datasets | Yes | Our evaluation is carried out i) on a variety of standard numeric domains from the International Planning Competition (IPC)12 and the LPRPG benchmark suite (Coles et al., 2013), ii) on numeric planning reformulations of Functional STRIPS problems (Franc es & Geffner, 2015) (F-STRIPS), and iii) on two sequential numeric domains called Sailing and Farmland introduced in our previous work (Scala et al., 2016a). |
| Dataset Splits | No | For the first set we have 11 instances; for the random variant we have 33 instances; for the reversed variant 11 instances. ... Instances generated for this domain scale on the number of boats (from 1 to 4) and the number of people to rescue (from 1 to 10). ... For this optimal setting, Counters instances consist of 2 up to 9 counters, for a total of 8 instances. The paper specifies instance counts for different problem variants and scaling factors for domains, but it does not describe specific train/test/validation splits for a dataset. The evaluation is performed on predefined sets of problem instances. |
| Hardware Specification | Yes | Our experiments were run under Ubuntu on an Intel(R) Xeon(R) CPU E3-1240 v3 @ 3.40GHz and 8Gb of Ram with a timeout of 1800 secs. per problem. |
| Software Dependencies | Yes | For solving the linear programs in the hgen hbd formulation we used CPLEX 12.6.3. |
| Experiment Setup | Yes | We employed such heuristics in a forward state space search planner guided by two different informed search algorithms: Greedy Best First Search (GBFS) for the satisficing setting, and A for the optimal setting. In the satisficing setting ties are broken to favour nodes with lower g-values; in the optimal setting ties are broken to favour nodes with higher g-values. The implementation is part of the ENHSP planning system... Our experiments were run under Ubuntu on an Intel(R) Xeon(R) CPU E3-1240 v3 @ 3.40GHz and 8Gb of Ram with a timeout of 1800 secs. per problem. In our experiments, all actions have unit cost, therefore the optimal plans are those with the shortest length. |