Research map / Multi-agent systems and games / Cooperative multi-agent RL
Revisiting Some Common Practices in Cooperative 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 |
|---|---|---|---|
| Use agent-specific policies instead of one shared policy | PG on Bridge · Bridge | reward: -0.64(0.02) -> -0.48(0.00) | Helped |
| Use agent-ID-conditioned policies rather than shared policies | MAPPO · SMAC selected maps | winning rate | Helped |
| Use agent-specific policy gradient instead of value decomposition | MARL policies · SMAC selected maps | winning rate | Helped |
| Use PG-ID or PG-Ind instead of CDS | GRF · GRF academy scenarios except PS | winning rate | Helped |
| Use PG-Ind rather than PG-ID | GRF policy gradient · GRF academy scenarios | winning rate | Mixed |
| Use PG-Ind rather than PG-ID on maps with many agents | SMAC policy gradient · SMAC maps with many agents | winning rate | Hurt |
| Use an auto-regressive policy rather than independent policies | 4-player permutation game · 4-player permutation game | policy entropy | Helped |
| Add multi-step optimization to auto-regressive policy learning (ablation) | PG-AR · Bridge state at bridge ends | policy entropy: 0.02(0.01) -> 0.74(0.03) | Helped |
| Randomize execution order during training (ablation) | PG-AR · Bridge state at bridge ends | policy entropy: 0.71(0.28) -> 0.74(0.03) | No clear effect |
| Add self-attention to the auto-regressive policy (ablation) | PG-AR · Bridge state at bridge ends | policy entropy: 0.60(0.44) -> 0.74(0.03) | Helped |
| Extend HAPPO with auto-regressive action conditioning | HAPPO · GRF academy scenarios | winning rate | Mixed |
| Use value decomposition in XOR instead of policy gradient | XOR game · XOR game | Q-function fit | Hurt |
Closest papers by meaning
- Benchmarking Multi-Agent Deep Reinforcement Learning Algorithms in Cooperative Tasks (NeurIPS 2021 · similarity 0.94)
- Towards a Standardised Performance Evaluation Protocol for Cooperative MARL (NeurIPS 2022 · similarity 0.92)
- Gradient-Protected Value Decomposition for Cooperative Multi-Agent Reinforcement Learning (AAAI 2026 · similarity 0.92)
- HyperMARL: Adaptive Hypernetworks for Multi-Agent RL (NeurIPS 2025 · similarity 0.91)
- Scalable Constrained Policy Optimization for Safe Multi-agent Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Learning Joint Behaviors with Large Variations (AAAI 2025 · similarity 0.90)
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.