This repository contains the official implementation of ADAPT (Auction-based Dynamic Action Priority Technique), a dynamic coordination framework for multi-agent reinforcement learning (MARL) designed to enhance coordination under partial observability and synchronous execution. ADAPT combines:

• Transformer-based observation encoding to extract compact and informative features from high-dimensional observations.
• Message generation with mutual information objectives to produce meaningful and informative messages for communication.
• Dynamic priority scheduling via distributed auctions to assign execution priorities based on real-time inter-agent dependencies.
• Autoregressive action inference to model a learned causal order for policy inference.
• Observation reconstruction to recover global observation embeddings from compact messages.

Experiments on StarCraft Multi-Agent Challenge v2 (SMACv2) and Google Research Football (GRF) show that ADAPT achieves significantly higher win rates and reduced communication overhead compared to state-of-the-art baselines. For example, on SMACv2's Terran_10_vs_11 map, ADAPT attains a win rate of 53.70%, surpassing CommFormer (31.47%) and SeqComm (30.41%). ADAPT also demonstrates strong generalization on GRF tasks. In terms of communication efficiency, ADAPT reduces message bytes per timestep by up to 51.54% compared to SeqComm and 3.77% compared to CommFormer.

This code is implemented based on https://github.com/marlbenchmark/on-policy, and the running instructions are similar to that in the original project.

pip install -r requirements.txt

Run the script

bash install_sc2.sh

Or you could install them manually to other path you like, just follow here: https://github.com/oxwhirl/smac.

Please following the instructios in https://github.com/google-research/football.

When your environment is ready, you could run shells in the "scripts" folder with algo="ADAPT". For example:

bash ./train_smacv2.sh  # run ADAPT on SMACv2

If you would like to change the configs of experiments, you could modify sh files or look for config.py for more details.

If you find this project helpful, please consider to cite the following paper:

@inproceedings{Xie2025ADAPT,
  author    = {Xie, Zaipeng and Qiao, Chentai and Yang, Nuo and Zhao, Yiming},
  title     = {ADAPT: Auction-Based Dynamic Prioritization for Multi-Agent Coordination},
  booktitle = {Proceedings of the 28th European Conference on Artificial Intelligence (ECAI 2025)},
  year      = {2025},
  address   = {Bologna, Italy},
  month     = {October 25--30},
  note      = {in press},
}