scalableMARL

Scalable Reinforcement Learning Policies for Multi-Agent Control

CD. Hsu, H. Jeong, GJ. Pappas, P. Chaudhari. "Scalable Reinforcement Learning Policies for Multi-Agent Control". IEEE International Conference on Intelligent Robots and Systems (IROS), Prague, Czech Republic, 2021.

Multi-Agent Reinforcement Learning method to learn scalable control polices for multi-agent target tracking.

• Author: Christopher Hsu
• Email: chsu8@seas.upenn.edu
• Affiliation:
  • Department of Electrical and Systems Engineering
  • GRASP Laboratory
  • @ University of Pennsylvania

Currently supports Python3.8 and is developed in Ubuntu 20.04

scalableMARL file structure

Within scalableMARL (highlighting the important files):

scalableMARL
    |___algos
        |___maTT                          #RL alg folder for the target tracking environment
            |___core                      #Self-Attention-based Model Architecture
            |___core_behavior             #Used for further evaluation (Ablation D.2.)
            |___dql                       #Soft Double Q-Learning
            |___evaluation                #Evaluation for Main Results
            |___evaluation_behavior       #Used for further evaluation (Ablation D.2.)
            |___modules                   #Self-Attention blocks
            |___replay_buffer             #RL replay buffer for sets
            |___run_script                #**Main run script to do training and evaluation
    |___envs
        |___maTTenv                       #multi-agent target tracking
            |___env
                |___setTracking_v0        #Standard environment (i.e. 4a4t tasks)
                |___setTracking_vGreedy   #Baseline Greedy Heuristic
                |___setTracking_vGru      #Experiment with Gru (Ablation D.3)
                |___setTracking_vkGreedy  #Experiment with Scalability and Heuristic Mask k=4 (Ablation D.1)
        |___run_ma_tracking               #Example scipt to run environment
    |___setup                             #set PYTHONPATH ($source setup)

• To setup scalableMARL, follow the instruction below.

Set up python environment for the scalableMARL repository

Install python3.8 (if it is not already installed)

#to check python version
python3 -V

sudo apt-get update
sudo apt-get install python3.8-dev

Set up environment (conda or virtualenv)

Set up with conda

conda env create -f environment.yml
conda activate scalableMARL
source setup

Set up virtualenv

Python virtual environments are used to isolate package installation from the system

Replace 'virtualenv name' with your choice of folder name

sudo apt-get install python3-venv 

python3 -m venv --system-site-packages ./'virtualenv name'
# Activate the environment for use, any commands now will be done within this venv
source ./'virtualenv name'/bin/activate

# To deactivate (in terminal, exit out of venv), do not use during setup
deactivate

Now that the virtualenv is activated, you can install packages that are isolated from your system

When the venv is activated, you can now install packages and run scripts

Install isolated packages in your venv

sudo apt-get install -y eog python3-tk python3-yaml python3-pip ssh git

#This command will auto install packages from requirements.txt
pip3 install --trusted-host pypi.python.org -r requirements.txt

Current workflow

Setup repos

# activate env
conda activate scalableMARL
# or
source ./'virtualenv name'/bin/activate
# change directory to scalableMARL
cd ./scalableMARL
# setup repo  ***important in order to set PYTHONPATH***
source setup

scalableMARL repo is ready to go

Running an algorithm

# its best to run from the scalableMARL folder so that logging and saving is consistent
cd ./scalableMARL
# run the alg
python3 algos/maTT/run_script.py

# you can run the alg with different argument parameters. See within run_script for more options.
# for example
python3 algos/maTT/run_script.py --seed 0 --logdir ./results/maPredPrey --epochs 40

To test, evaluate, and render()

# for a general example 
python3 algos/maTT/run_script.py --mode test --render 1 --log_dir ./results/maTT/setTracking-v0_123456789/seed_0/ --nb_test_eps 50
# for a saved policy in saved_results
python3 algos/maTT/run_script.py --mode test --render 1 --log_dir ./saved_results/maTT/setTracking-v0_123456789/seed_0/

To see training curves

tensorboard --logdir ./results/maTT/setTracking-v0_123456789/

Citing scalableMARL

If you reference or use scalableMARL in your research, please cite:

@misc{hsu2021scalable,
      title={Scalable Reinforcement Learning Policies for Multi-Agent Control}, 
      author={Christopher D. Hsu and Heejin Jeong and George J. Pappas and Pratik Chaudhari},
      year={2021},
      eprint={2011.08055},
      archivePrefix={arXiv},
      primaryClass={cs.MA}
}