PDDLEGO: Iterative Planning in Textual Environments

This README provides usage to replicate the methods described in the paper.

First, install TextWroldExpress.

To run experiment on either Coin Collector or Cooking World:

python run_coin.py|run_cooking.py --model MODEL --method METHOD --split SPLIT [--det] [--oc]

• model: OpenAI model names such as gpt-3.5-turbo-1106 or gpt-4-1106-preview
• method: one of direct (Action-gen) or pddl (PDDL-gen and PDDL-edit)
• split: one of dev or test, or a singular number to specify a specifc example of that index
• det: must be used when method is pddl; if specified, use PDDL-edit, else PDDL-gen
• oc: whether to overwrite existing cache; if not specified, will use the previous cache

Citation

@inproceedings{zhang-etal-2024-pddlego,
    title = "PDDLEGO: Iterative Planning in Textual Environments",
    author = "Zhang, Li and Jansen, Peter and Zhang, Tianyi and Clark, Peter and Callison-Burch, Chris and Tandon, Niket",
    booktitle = "Proceedings of the 13th Joint Conference on Lexical and Computational Semantics (*SEM 2024)",
    month = jun,
    year = "2024",
    address = "Mexico City, Mexico",
    publisher = "Association for Computational Linguistics"
}