ABCS Flocking: A Leader-Follower Collision Free UAV Flocking System with Multi-Agent Reinforcement Learning

This is the Attention Based Cucker-Smale Flocking algorithm implementation on Multi-Agent Particle Environment(MPE), the corresponding paper is ABCS Flocking: A Leader-Follower Collision Free UAV Flocking System with Multi-Agent Reinforcement Learning

The MADDPG part is come from: MADDPG

Requirements

• python=3.6.5
• Multi-Agent Particle Environment(MPE)
• torch=1.1.0

Or download the python environment directly: LG-CS.zip

Extract code: MARL

Complie Cython Code

Before running the code, please complie the environment code:

cd ./envs
python setup.py build_ext --inplace --force

Training Agents

Running the main.py, the agents will learn from the flocking scenario：

python main.py --n-agents=5 --evaluate-episodes=256

If you want to adjust the parameters, please see the ./common/arguments.py for more details.

Testing Agents

python main.py --n-agents=5 --evaluate-episodes=10 --evaluate=True

Display Results

After data collection:

python display.py