Research map / Multi-agent systems and games / Cooperative multi-agent RL
CM3: Cooperative Multi-goal Multi-stage Multi-agent Reinforcement Learning
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 |
|---|---|---|---|
| Train individual goal attainment first, then initialize multi-agent learning from that stage | CM3 · Cooperative navigation | episodes to convergence: more than 15k episodes faster than IAC | Helped |
| Use the two-stage curriculum and function augmentation instead of training the full Stage 2 architecture directly (ablation) | CM3 · Cooperative navigation, SUMO, and Checkers | learning speed and stability | Helped |
| Use a credit function for action-goal credit assignment instead of the QV global advantage (ablation) | CM3 · Cooperative navigation, SUMO, and Checkers | cooperative learning stability | Helped |
| Use CM3 rather than COMA | Multi-goal multi-agent learning · Cooperative navigation and SUMO | learning speed and final performance | Helped |
| Use CM3 rather than IAC | Multi-goal multi-agent learning · Cooperative navigation and SUMO | learning speed | Helped |
| Use CM3 rather than QMIX | Multi-goal multi-agent learning · Four of five evaluated environments | performance: four out of five | Helped |
| Use CM3 rather than QMIX or COMA | Multi-goal multi-agent learning · SUMO | performance | No clear effect |
| Use CM3 rather than COMA or QMIX | Multi-goal multi-agent learning · Checkers | episodes to convergence: 10k episodes faster than COMA and QMIX | Helped |
| Train with two agents and deploy in heavier traffic | CM3 · SUMO heavy traffic | sum reward | Helped |
| Use off-policy training with a persistent replay buffer | CM3 · SUMO and Checkers | performance | Helped |
Closest papers by meaning
- Multi-agent cooperation through learning-aware policy gradients (ICLR 2025 · similarity 0.88)
- Cross-environment Cooperation Enables Zero-shot Multi-agent Coordination (ICML 2025 · similarity 0.88)
- LAGMA: LAtent Goal-guided Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.88)
- POGEMA: A Benchmark Platform for Cooperative Multi-Agent Pathfinding (ICLR 2025 · similarity 0.88)
- Learning Joint Behaviors with Large Variations (AAAI 2025 · similarity 0.87)
- The Horizon Threshold in Cooperative Multi-Agent Reward-Free Exploration (NeurIPS 2026 · similarity 0.87)
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.