MA-LAMA: Exploiting the Multi-Agent Nature of Temporal Planning Problems

Authors: J. Caballero Testรณn, Maria D. R-Moreno

JAIR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 7, we conduct a two-part empirical evaluation of MA-LAMA. First, we evaluate MA-LAMA s coverage and plan quality performance in different domains. Subsequently, we analyze its search efficiency by comparing runtime with the quality of the generated plans in complex scenarios.
Researcher Affiliation Academia J. CABALLERO TESTร“N , Universidad de Alcalรก, ISG, EPS, Spain MARIA D. R-MORENO, Universidad de Alcalรก, ISG, EPS, Spain and TNO, IAS, The Netherlands
Pseudocode Yes Algorithm 1 Agent Decomposition (AD) Input: MAP temporal task, ๐‘’๐‘€๐‘ƒ๐‘‡ ๐‘‰, ๐ผ,๐บ,๐‘‚, ๐‘, ๐‘€ , where ๐‘‰is the multi-valued variables set, ๐ผis the initial state, ๐บis the goals set, ๐‘‚is the operators set, ๐‘is the multi-valued numeric variables set and ๐‘€is the metric.
Open Source Code No No explicit statement about code release or repository link found.
Open Datasets Yes We test the performance of MA-LAMA through a MA temporal domains test bench, which comprises the MA temporal Exploration domain and all IPC temporal domains [12] that present MA features. ... [12] A. Coles, C. Coles, M. Martinez, and P. Sidiropoulos. 2018. International planning competition 2018 temporal tracks. https://ipc2018-te mporal.bitbucket.io/. Accessed: 2022-09-01. (2018).
Dataset Splits No The paper refers to planning problems from IPC temporal domains as test cases (e.g., 'Taxis (20)', 'Rovers (20)') but does not specify training/test/validation splits in the traditional machine learning sense for these problem sets. The entire set of problems is used for evaluation.
Hardware Specification No All executions are limited to 10 minutes and 4GB of RAM.
Software Dependencies No The paper mentions several planners and languages used (e.g., LAMA [45], PDDL2.1 [20]), but does not provide specific version numbers for any of them.
Experiment Setup Yes All executions are limited to 10 minutes and 4GB of RAM. ... The metric used is a weighted combination of battery usage and mission risk, similar to the one defined in Figure 3. ... The metric aims to minimize the overall sum, favoring UAV operations by assigning them lower penalty values compared to ROVER actions.