Reshaping Diverse Planning
Authors: Michael Katz, Shirin Sohrabi9892-9899
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments show that the proposed approach significantly improves over the best performing existing planners in terms of coverage, the overall solution quality, and the overall diversity according to various diversity metrics. |
| Researcher Affiliation | Industry | Michael Katz, Shirin Sohrabi IBM T.J. Watson Research Center 1101 Kitchawan Rd, Yorktown Heights, NY 10598, USA |
| Pseudocode | Yes | Algorithm 1 Iterative diverse planning scheme. Input: Planning task Π, number of diverse plans k, number of total plans for search phase K, diversity metric D P Π Π while |P| < K do π some solution to Π P P {π | π is symmetric to π} X π P Xπ Π Π X according to Definition 4 end while return choose k diverse plans from P, according to D |
| Open Source Code | Yes | Our planners, Forbid Iterative (FI) diverse planners are publicly available as part of the collection of Forbid Iterative planners (Katz, Sohrabi, and Udrea 2019). Further, we have implemented an external component, that given a set of plans and a metric returns the score of the set under that metric and made the code publicly available (Katz and Sohrabi 2019). |
| Open Datasets | Yes | To compare to all selected existing planners, we restrict our benchmark set to STRIPS domains with uniform action costs from the International Planning Competitions (IPC). This results in 1276 tasks in 40 domains. |
| Dataset Splits | No | The paper uses benchmark tasks from the International Planning Competitions (IPC) which are typically used for evaluation without traditional train/validation splits. It does not provide specific details on how data was split for training or validation. |
| Hardware Specification | Yes | The experiments were performed on Intel(R) Xeon(R) CPU E7-8837 @2.67GHz machines, with time and memory limits of 30min and 2GB, respectively. |
| Software Dependencies | No | The paper mentions using the "Fast Downward planning system (Helmert 2006)", "MERWIN (Katz et al. 2018a)", and the "CPLEX solver". While these are specific software names, no specific version numbers are provided for Fast Downward, MERWIN, or CPLEX. |
| Experiment Setup | Yes | We are restricting the number of found plans to 1000. For smaller k values (k = 5, 10), there is a clear advantage to our approach, for all tested bounds on Dmma. The solution is obtained by solving the binary linear program, as described in Section 4.2 with the CPLEX solver in its default configuration. |