Multi agent reinforcement learning: PyTorch implementations of several algorithms for Multi Agent domains
- MADDPG extends the DDPG algorithm to multi-agent settings by using centralized learning and decentralized execution. Each agent has a centralized critic that evaluates all agents' actions, while having a decentralized actor for decision-making.
- Paper: Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments
Demo | Results |
---|---|
git clone git@github.com:mobinajamali/MARL.git
pip install -r requirements.txt
Shout out to Phil Tabor for his brilliant courses!