Research map / Multi-agent systems and games / Cooperative multi-agent RL
FACMAC: Factored Multi-Agent Centralised Policy Gradients
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| FACMAC (factored critic + centralised policy gradient) instead of MADDPG | multi-agent actor-critic · Continuous Predator-Prey and MAMuJoCo | mean episode return | Helped |
| FACMAC instead of MADDPG | multi-agent actor-critic · SMAC (6 maps) | test win rate | Mixed |
| FACMAC instead of QMIX (same factorisation) | SMAC · SMAC MMM, bane_vs_bane, MMM2, 27m_vs_30m | test win rate | Helped |
| FACMAC instead of COMIX | continuous MARL · Continuous Predator-Prey and MAMuJoCo | mean episode return | No clear effect |
| factor the critic (FACMAC without CPG vs MADDPG) (ablation) | MADDPG · SMAC 2c_vs_64zg, MMM2 | test win rate | Helped |
| add centralised policy gradient (ablation) | FACMAC · SMAC MMM2 | test win rate | Helped |
| add centralised policy gradient (ablation) | FACMAC and MADDPG · ManyAgent Swimmer 2 agents | mean episode return | No clear effect |
| add centralised policy gradient (ablation) | FACMAC · ManyAgent Swimmer 10 agents | mean episode return | Helped |
| add centralised policy gradient (ablation) | MADDPG · Continuous Matrix Game | mean test return | Helped |
| replace nonlinear mixing with VDN sum (FACMAC-vdn) (ablation) | FACMAC · MAMuJoCo Humanoid, ManyAgent Swimmer | mean episode return | Hurt |
| replace nonlinear mixing with VDN sum (FACMAC-vdn) (ablation) | FACMAC · Continuous Predator-Prey many agents | mean episode return | Helped |
| use nonmonotonic factorised critic (ablation) | FACMAC · nonmonotonic matrix game / predator-prey | return | Helped |
Closest papers by meaning
- Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments (NeurIPS 2017 · similarity 0.90)
- QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning (ICML 2018 · similarity 0.90)
- More Centralized Training, Still Decentralized Execution: Multi-Agent Conditional Policy Factorization (ICLR 2023 · similarity 0.90)
- Revisiting Some Common Practices in Cooperative Multi-Agent Reinforcement Learning (ICML 2022 · similarity 0.89)
- Learning Implicit Credit Assignment for Cooperative Multi-Agent Reinforcement Learning (NeurIPS 2020 · similarity 0.89)
- Counterfactual Multi-Agent Policy Gradients (AAAI 2018 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.