LEMURS: LEarning distributed MUlti-Robot interactionS

This repository provides the code for our paper "LEMURS: Learning Distributed Multi-robot Interactions". Please check out our project website for more details: https://eduardosebastianrodriguez.github.io/LEMURS/.

Dependencies

Our code is tested with Ubuntu 20.04 and Python 3.8. It depends on the following Python packages:

torchdiffeq 0.2.3

torch 1.12.1

numpy 1.17.4

matplotlib 3.5.2

Demo with fixed swapping problem

Run python ./FixedSwapping/Demo.py to check the learned control policy for the fixed swapping problem. It uses the pretrained model stored in ./FixedSwapping/FS4_42_learn_system_LEMURS.pth

To generate the training and evaluation datasets, run python ./FixedSwapping/DatasetGenerator.py, choosing the following arguments as you prefer:

--numTrain : number of training trajectories. It is 20000 by default.

--numTests: number of testing trajectories. It is 20000 by default.

--numSamples: number of samples per trajectory. It is 5 by default.

--seed: set random seed. It is 42 by default.

--numAgents: number of robots used as ground-truth. It is 4 by default.

To train you own model, run python ./FixedSwapping/TrainingLEMURS.py, choosing the following arguments as you prefer:

--numTrain : number of training trajectories in the dataset. It is 20000 by default.

--numTests: number of testing trajectories in the dataset. It is 20000 by default.

--numSamples: number of samples per trajectory in the dataset. It is 5 by default.

--seed_data: random seed used to generate the dataset. It is 42 by default.

--seed_train: set random seed for the training. It is 42 by default.

--numAgents: number of robots used in the dataset. It is 4 by default.

To train the MLP, GNN and GNNSA models, run python ./FixedSwapping/TrainingMLP.py, python ./FixedSwapping/TrainingGNN.py and python ./FixedSwapping/TrainingGNNSA.py respectively, using the same arguments detailed above.

The files python ./FixedSwapping/Evaluation.py, python ./FixedSwapping/Training.py, python ./FixedSwapping/Scalability.py evaluate LEMURS, compare the different models and check the scalability of LEMURS respectively. Regarding python ./FixedSwapping/Evaluation.py, choose the arguments according to the ones used for generating the datasets and training the models.

Demo with time-varying swapping problem

Run python ./TimeVaryingSwapping/Demo.py to check the learned control policy for the time-varying swapping problem. It uses the pretrained model stored in ./TimeVaryingSwapping/FS4_42_learn_system_LEMURS.pth

To generate the training and evaluation datasets, run python ./TimeVaryingSwapping/DatasetGenerator.py, choosing the following arguments as you prefer:

--numTrain : number of training trajectories. It is 20000 by default.

--numTests: number of testing trajectories. It is 20000 by default.

--numSamples: number of samples per trajectory. It is 5 by default.

--seed: set random seed. It is 42 by default.

--numAgents: number of robots used as ground-truth. It is 4 by default.

To train you own model, run python ./TimeVaryingSwapping/TrainingLEMURS.py, choosing the following arguments as you prefer:

--numTrain : number of training trajectories in the dataset. It is 20000 by default.

--numTests: number of testing trajectories in the dataset. It is 20000 by default.

--numSamples: number of samples per trajectory in the dataset. It is 5 by default.

--seed_data: random seed used to generate the dataset. It is 42 by default.

--seed_train: set random seed for the training. It is 42 by default.

--numAgents: number of robots used in the dataset. It is 4 by default.

To train the MLP, GNN and GNNSA models, run python ./TimeVaryingSwapping/TrainingMLP.py, python ./TimeVaryingSwapping/TrainingGNN.py and python ./TimeVaryingSwapping/TrainingGNNSA.py respectively, using the same arguments detailed above.

The files python ./TimeVaryingSwapping/Evaluation.py, python ./TimeVaryingSwapping/Training.py, python ./TimeVaryingSwapping/Scalability.py evaluate LEMURS, compare the different models and check the scalability of LEMURS respectively. Regarding python ./TimeVaryingSwapping/Evaluation.py, choose the arguments according to the ones used for generating the datasets and training the models.

Demo with flocking problem

Run python ./Flocking/Demo.py to check the learned control policy for the flocking problem. It uses the pretrained model stored in ./Flocking/FS4_42_learn_system_LEMURS.pth

To generate the training and evaluation datasets, run python ./Flocking/DatasetGenerator.py, choosing the following arguments as you prefer:

--numTrain : number of training trajectories. It is 20000 by default.

--numTests: number of testing trajectories. It is 20000 by default.

--numSamples: number of samples per trajectory. It is 5 by default.

--seed: set random seed. It is 42 by default.

--numAgents: number of robots used as ground-truth. It is 4 by default.

To train you own model, run python ./Flocking/TrainingLEMURS.py, choosing the following arguments as you prefer:

--numTrain : number of training trajectories in the dataset. It is 20000 by default.

--numTests: number of testing trajectories in the dataset. It is 20000 by default.

--numSamples: number of samples per trajectory in the dataset. It is 5 by default.

--seed_data: random seed used to generate the dataset. It is 42 by default.

--seed_train: set random seed for the training. It is 42 by default.

--numAgents: number of robots used in the dataset. It is 4 by default.

To train the MLP, GNN and GNNSA models, run python ./Flocking/TrainingMLP.py, python ./Flocking/TrainingGNN.py and python ./Flocking/TrainingGNNSA.py respectively, using the same arguments detailed above.

The files python ./Flocking/Evaluation.py, python ./Flocking/Training.py, python ./Flocking/Scalability.py evaluate LEMURS, compare the different models and check the scalability of LEMURS respectively. Regarding python ./Flocking/Evaluation.py, choose the arguments according to the ones used for generating the datasets and training the models.

Citation

If you find our papers/code useful for your research, please cite our work as follows.

E. Sebastian, T. Duong, N. Atanasov, E. Montijano, C. Sagues. LEMURS: Learning Distributed Multi-robot Interactions. IEEE ICRA 2023

@article{sebastian22LEMURS,
author = {Eduardo Sebasti\'{a}n AND Thai Duong AND Nikolay Atanasov AND Eduardo Montijano AND Carlos Sag\"{u}\'{e}s},
title = {{LEMURS: Learning Distributed Multi-robot Interactions}},
journal = {IEEE International Conference on Robotics and Automation},
year = {2023}
}