Research map / Multi-agent systems and games / Cooperative multi-agent RL
Counterfactual Conservative Q Learning for Offline Multi-agent Reinforcement Learning
Jianzhun Shao, Yun Qu, Chen Chen, Hongchang Zhang, Xiangyang Ji
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| use counterfactual per-agent conservatism instead of joint-action CQL (MACQL) | CFCQL · MPE Cooperative Navigation Expert | normalized score: 12.2±31 -> 112±4 | Improved |
| counterfactual conservatism instead of MACQL | CFCQL · SMAC 5m_vs_6m expert | test win rate: 0.01±0.01 -> 0.84±0.03 | Improved |
| counterfactual conservatism instead of MACQL | CFCQL · SMAC 6h_vs_8z expert | test win rate: 0.00±0.00 -> 0.7±0.06 | Improved |
| increase the number of agents n (ablation) | MACQL vs CFCQL · Equal_Line | performance ratio to behaviour policy | Mixed |
| CFCQL vs OMAR on random-quality data | CFCQL · MaMuJoCo HalfCheetah Random | normalized score: 13.5±7.0 -> 39.7±4.0 | Improved |
| CFCQL vs IQL on random data | CFCQL · MPE Cooperative Navigation Random | normalized score: 5.5±1.1 -> 62.2±8.1 | Improved |
| CFCQL vs OMAR on expert data | CFCQL · MPE Cooperative Navigation Expert | normalized score: 114.9±2.6 -> 112±4 | Hurt |
| CFCQL vs best baseline on expert data | CFCQL · MaMuJoCo HalfCheetah Expert | normalized score: 115.6±4.2 -> 118.5±4.9 | No clear effect |
| CFCQL vs MAICQ on mixed data | CFCQL · SMAC 2s3z mixed | test win rate: 0.85±0.07 -> 0.84±0.09 | No clear effect |
| vary temperature coefficient tau (ablation) | CFCQL · SMAC 6h_vs_8z | test win rate | No clear effect |
| tune tau beyond tau=0 (ablation) | CFCQL · Multi-agent Particle Environment | normalized score | Improved |
| replace counterfactual policy improvement with MADDPG policy improvement (ablation) | CFCQL · MaMuJoCo | normalized score | Hurt |
Closest papers by meaning
- Peng's Q($\lambda$) for Conservative Value Estimation in Offline Reinforcement Learning (ICLR 2026 · similarity 0.93)
- A Perspective of Q-value Estimation on Offline-to-Online Reinforcement Learning (AAAI 2024 · similarity 0.90)
- On the Complexity of Offline Reinforcement Learning with Q*-Approximation and Partial Coverage (NeurIPS 2026 · similarity 0.89)
- Exclusively Penalized Q-learning for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- ComaDICE: Offline Cooperative Multi-Agent Reinforcement Learning with Stationary Distribution Shift Regularization (ICLR 2025 · similarity 0.88)
- Compositional Conservatism: A Transductive Approach in Offline Reinforcement Learning (ICLR 2024 · similarity 0.88)
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.
