PMAC: Personalized Multi-Agent Communication

Authors: Xiangrui Meng, Ying Tan

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we show the strength of personalized communication in a variety of cooperative scenarios. Our approach exhibits competitive performance compared to existing methods while maintaining notable computational efficiency. We empirically evaluate PMAC on a variety of multi-agent cooperative tasks which are hard-level Traffic Junction (hard TJ) (Sukhbaatar, Szlam, and Fergus 2016), Cooperative Navigation (CN), and Predator-Prey (PP) in Multi-agent Particle Environment (MPE) (Mordatch and Abbeel 2017; Lowe et al. 2017), and Google Research Football (GRF) (Kurach et al. 2020).
Researcher Affiliation Academia Xiangrui Meng1,2, Ying Tan1,2,3,4 * 1 School of Intelligence Science and Technology, Peking University 2 Key Laboratory of Machine Perceptron (MOE), Peking University 3 Institute for Artificial Intelligence, Peking University 4 National Key Laboratory of General Artificial Intelligence mxxxr@stu.pku.edu.cn, ytan@pku.edu.cn
Pseudocode No The paper does not contain a formally labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code No The paper does not provide any statement or link regarding the public availability of the source code for the described methodology.
Open Datasets Yes We empirically evaluate PMAC on a variety of multi-agent cooperative tasks which are hard-level Traffic Junction (hard TJ) (Sukhbaatar, Szlam, and Fergus 2016), Cooperative Navigation (CN), and Predator-Prey (PP) in Multi-agent Particle Environment (MPE) (Mordatch and Abbeel 2017; Lowe et al. 2017), and Google Research Football (GRF) (Kurach et al. 2020).
Dataset Splits No The paper mentions running experiments for a certain number of episodes and over multiple runs (e.g., '1,000,000 episodes', 'standard deviations over 3 runs'), but it does not specify explicit training, validation, or test dataset splits in terms of percentages or sample counts for the environments used.
Hardware Specification No The paper does not provide any specific hardware details such as GPU or CPU models, memory, or cloud instance specifications used for running the experiments.
Software Dependencies No The paper does not provide specific software dependency details, such as library names with version numbers (e.g., Python 3.8, PyTorch 1.9).
Experiment Setup Yes We use the hard level TJ task on a 18 × 18 grid (see Fig. 4(a)). The total number of cars at any given time is limited to N = 20. We use actor-critic (Sutton and Barto 2018) to train our model. PMAC demonstrates faster convergence and achieves higher rewards when η = 0.001.