Pytorch implementation of the paper "Cooperative and Asynchronous Transformer-based Mission Planning for Heterogeneous Teams of Mobile Robots".

Training

You could start training with by running sh train_gridworld.sh in directory matsar/scripts.

Evaluation

Similar to training, you could run sh render_gridworld.sh in directory matsar/scripts to start evaluation. Remember to set up your path to the cooresponding model, correct hyperparameters and related evaluation parameters.

Results

The performance of a model trained on a $20\times 20$ map with one rescuer agent, one explorer agent, and one randomly assigned agent per episode is visualized on the three tasks with different map sizes and team compositions, as described in the paper.

Task 1: One rescuer and one explorer agent in a $16 \times 16$ map.	Task 2: One rescuer, one explorer, and one randomly assigned agent in a $20 \times 20$ map.	Task 3: Two rescuers, four explorers, and two randomly assigned agents in a $32\times 32$ map.

Citation

The preprint is now available on arXiv:

@misc{farjadnasab2024cooperative,
    title={Cooperative and Asynchronous Transformer-based Mission Planning for Heterogeneous Teams of Mobile Robots},
    author={Milad Farjadnasab and Shahin Sirouspour},
    year={2024},
    eprint={2410.06372},
    archivePrefix={arXiv},
    primaryClass={cs.RO}
}