PORTAL: Automatic Curricula Generation for Multiagent Reinforcement Learning

Authors: Jizhou Wu, Jianye Hao, Tianpei Yang, Xiaotian Hao, Yan Zheng, Weixun Wang, Matthew E. Taylor

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that PORTAL can train agents to master extremely hard cooperative tasks, which can not be achieved with previous state-of-the-art MARL algorithms. In this section, we study the following research questions (RQs) via comprehensive experiments.
Researcher Affiliation Collaboration 1College of Intelligence and Computing, Tianjin University 2University of Alberta and Alberta Machine Intelligence Institute 3Netease Fuxi AI Lab
Pseudocode Yes Algorithm 1: PORTAL
Open Source Code Yes Code and appendix: https://github.com/TJU-DRLLAB/transfer-and-multi-task-reinforcement-learning
Open Datasets No The paper mentions using "Starcraft Multi Agent Challenge (SMAC) (Samvelyan et al. 2019)" as the benchmark, but does not provide concrete access information like a link, DOI, or specific repository for the dataset itself. While SMAC is a well-known benchmark, the paper doesn't explicitly state its public availability with a direct source.
Dataset Splits No The paper mentions "test win rate" but does not explicitly specify training, validation, or test splits by percentages, counts, or references to predefined splits within the SMAC environment. It describes task series and scenarios but not how the data within those tasks is split for training/validation/testing.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, memory, or cloud instance types used for running experiments.
Software Dependencies No The paper mentions using "HPN-VDN (Hao et al. 2022)" as a backbone model and refers to other MARL algorithms, but does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow, specific libraries, or their versions).
Experiment Setup No The paper states "detailed model structure and hyperparameter settings are in Appendix B" but does not include these details in the main text. The main text describes the experimental setup in terms of environments and baselines, but lacks concrete hyperparameter values or training configurations.