Learning General Planning Policies from Small Examples Without Supervision

Authors: Guillem Francès, Blai Bonet, Hector Geffner11801-11808

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We implemented the proposed approach in a C++/Python system called D2L and evaluated it on several problems. Results Table 1 provides an overview of the execution of D2L over all generalized domains.
Researcher Affiliation Academia Universitat Pompeu Fabra, Barcelona, Spain
Pseudocode No The paper describes its methods in prose and mathematical formulations but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes Source code and benchmarks are available online2 and archived in Zenodo (Franc es, Bonet, and Geffner 2021b). 2https://github.com/rleap-project/d2l.
Open Datasets Yes The domains include all problems with simple goals from (Bonet, Franc es, and Geffner 2019), e.g. clearing a block or stacking two blocks in Blocksworld, plus standard PDDL domains such as Gripper, Spanner, Miconic, Visitall and Blocksworld. We have used the benchmark distribution in https://github. com/aibasel/downward-benchmarks.
Dataset Splits No The paper mentions 'training instances' and 'test instances' but does not explicitly describe a 'validation' dataset split or process.
Hardware Specification Yes All experiments were run on an Intel i7-8700 CPU@3.2GHz with a 16 GB memory limit.
Software Dependencies No The paper mentions 'Open-WBO Weighted Max SAT solver', 'C++/Python system', and 'modified version of the Pyperplan planner' but does not specify their version numbers.
Experiment Setup Yes In all the experiments, we use δ = 2 and k F = 8, except in Delivery, where k F = 9 is required to find a policy.