Temporal Numeric Planning with Patterns
Authors: Matteo Cardellini, Enrico Giunchiglia
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To test the effectiveness of our approach, we compare our planner with all publicly available temporal planners (both symbolic and based on search) on 10 temporal domains with required concurrency (Cushing et al. 2007). The results highlight the strong performances of our planner, which achieved the highest coverage (i.e., number of solved problems) in 9 out of 10 domains, while the second-best planner had the highest coverage in 4 domains. Additionally, compared to the other symbolic planners, our system is able to find a valid plan with a lower bound on all the problems. [...] Table 1: Comparative analysis. |
| Researcher Affiliation | Academia | DIBRIS, Universit a degli Studi di Genova, Italy EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology and encoding process in detail using mathematical notation and textual explanations across sections like "The Pattern State Encoding" and "The Pattern Time Encoding", but it does not present any formal pseudocode or algorithm blocks. |
| Open Source Code | Yes | PATTYT and the PDDL 2.1 and ANML encoding of the new domains are available at https://github.com/matteocarde/patty . |
| Open Datasets | Yes | Table 1 presents the experimental analysis on the CUSHING domain (the only domain with required concurrency in the last International Planning Competition (IPC) with a temporal track (Coles et al. 2018)), all the domains and problems presented in (Panjkovic and Micheli 2023) (last five), and four new domains covering different types of required concurrency specified in (Cushing et al. 2007). |
| Dataset Splits | No | The experiments have been run using the same settings used in the Numeric/Agile Track of the last IPC, with 20 problems per domain and a time limit of 5 minutes. The paper mentions the number of problems per domain but does not provide specific training/test/validation splits or methodologies for data partitioning. |
| Hardware Specification | Yes | Analyses have been run on an Intel Xeon Platinum 8000 3.1GHz with 8 GB of RAM. |
| Software Dependencies | Yes | our system PATTYT implemented by modifying the planner PATTY (Cardellini, Giunchiglia, and Maratea 2024) and using the SMT-solver Z3 v4.8.7 (De Moura and Bjørner 2008); |
| Experiment Setup | Yes | The experiments have been run using the same settings used in the Numeric/Agile Track of the last IPC, with 20 problems per domain and a time limit of 5 minutes. [...] ANMLSMT and OPTIC have been set in order to return the first valid plan they find. |