Research map / Multi-agent systems and games / Cooperative multi-agent RL
A Structured Prediction Approach for Generalization in Cooperative Multi-Agent Reinforcement Learning
Nicolas Carion, Nicolas Usunier, Gabriel Synnaeve, Alessandro Lazaric
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 quadratic rather than greedy assignment with the direct scoring model (ablation) | Search and Rescue · 8×15, trained on 2×4 | average steps: 11.06 -> 7.85 | Improved |
| Use quadratic rather than linear-program assignment with the direct scoring model (ablation) | Search and Rescue · 8×15, trained on 2×4 | average steps: 9.71 -> 7.85 | Improved |
| Use direct scoring instead of positional embeddings (ablation) | Quad assignment · Search and Rescue, train 2×4, test 8×15 | average steps: 10.3 -> 7.85 | Improved |
| Use quadratic assignment instead of LP assignment (ablation) | StarCraft Wraiths · w60v67, trained on w15v17 | win rate: 0.33 -> 0.72 | Improved |
| Use quadratic assignment instead of greedy assignment (ablation) | StarCraft Wraiths · w60v67, trained on w15v17 | win rate: 0.13 -> 0.72 | Improved |
| Use quadratic assignment instead of LP assignment (ablation) | StarCraft Marines · m30v30, trained on m10v10 | win rate: 0.74 -> 0.06 | Hurt |
| Use quadratic assignment instead of LP assignment (ablation) | StarCraft Zergling-Hydralisk · zh10v12, trained on zh10v10 | win rate: 0.06 -> 0.01 | Hurt |
| Use positional-embedding scoring rather than direct scoring (ablation) | AMax assignment · Search and Rescue, train 2×4, test 8×15 | average steps: 11.06 -> 15.8 | Hurt |
| Add capacity constraints to assignment (ablation) | Search and Rescue · 8×15, trained on 2×4 | average steps: 11.06 -> 9.71 | Improved |
| Add quadratic task-task coordination scores to LP assignment (ablation) | StarCraft Wraiths · w15v17, trained on w15v17 | win rate: 0.53 -> 0.89 | Improved |
Closest papers by meaning
- SMACv2: An Improved Benchmark for Cooperative Multi-Agent Reinforcement Learning (NeurIPS 2023 · similarity 0.88)
- Multi-agent In-context Coordination via Decentralized Memory Retrieval (AAAI 2026 · similarity 0.88)
- Locality Matters: A Scalable Value Decomposition Approach for Cooperative Multi-Agent Reinforcement Learning (AAAI 2022 · similarity 0.88)
- Discovering Generalizable Multi-agent Coordination Skills from Multi-task Offline Data (ICLR 2023 · similarity 0.88)
- Cross-environment Cooperation Enables Zero-shot Multi-agent Coordination (ICML 2025 · similarity 0.87)
- Adaptive Collaboration with Humans: Metacognitive Policy Optimization for Multi-Agent LLMs with Continual Learning (ICLR 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.
