Generalising Planning Environment Redesign

Authors: Alberto Pozanco, Ramon Fraga Pereira, Daniel Borrajo

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments over a set of environment redesign benchmarks show that our general approach outperforms existing approaches when using well-known metrics, such as facilitating the recognition of goals, as well as its effectiveness when solving environment redesign tasks that optimise a novel set of different metrics. We now present the experiments carried out to evaluate GER.
Researcher Affiliation Collaboration Alberto Pozanco1*, Ramon Fraga Pereira2*, and Daniel Borrajo1 1J.P. Morgan AI Research 2University of Manchester, UK
Pseudocode Yes Algorithm 1: GER: A General Environment Redesign Approach
Open Source Code Yes Benchmarks and GER s code are available on Git Hub1. 1https://github.com/ramonpereira/general-environment-redesign
Open Datasets Yes We have created a benchmark set that contains 300 planning environment problems equally split across the five well-known domains: BLOCKS words, DEPOTS, GRID, IPC-GRID, and LOGISTICS. The environments are encoded in PDDL (Planning Domain Definition Language) (Mc Dermott et al. 1998). Benchmarks and GER s code are available on Git Hub1. 1https://github.com/ramonpereira/general-environment-redesign
Dataset Splits No The paper describes creating a benchmark set and evaluating performance, but does not specify explicit training, validation, or test dataset splits.
Hardware Specification Yes We have run all experiments using 4v CPU AMD EPYC 7R13 Processor 2.95GHz with 32GB of RAM
Software Dependencies No The paper mentions PDDL and SYM-K (von Tschammer, Mattm uller, and Speck 2022) but does not specify version numbers for these or other software dependencies.
Experiment Setup Yes We run GER with C = {time limit = 900s or memory limit = 4GB}. We used the same stopping condition C for both GER and GRD-LS. We also set a limit of 1, 000 plans to prevent disk overflows and avoid GER spending all the time computing the plan-library in redesign problems with a large number of optimal plans.