Flocking with Selective Evolutionary Multi-Agent Reinforcement Learning (EMARL) Algorithm

This is the Selective Evolutionary Multi-Agent Reinforcement Learning （SEMARL） algorithm implementation on Pytorch version, the corresponding paper is Cooperation and Competition: Flocking with Evolutionary Multi-Agent Reinforcement Learning (accepted by 29th ICONIP, conference version) & Semarl: Selective Evolutionary Multi-Agent Reinforcement Learning for Improving Cooperative Flocking with Competition (submitted, journal version)

Algorithms

• SEMARL (Selective Evolutionary Multi-Agent Reinforcement Learning, proposed)
• MADDPG (Multi-Agent Deep Deterministic Policy Gradient)
• COMA (Counterfactual Multi-Agent Policy Gradient)
• IQL (Independent Q-Learning (with DNNs))
• SQDDPG (Shapely Q-value Deep Deterministic Policy Gradient, not realized)

Requirements

• python=3.8.5
• torch>=1.13.1

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

Training Agents with SEMARL

If the python environment LG-CS is loaded, using follow instruction to train 15 agents (5 are senior agents, 5 are junior agents):

python main.py --n=15 --n-senior=5 --n-junior=5