Research map / Multi-agent systems and games / Cooperative multi-agent RL
Learning Decentralized LLM Collaboration with Multi-Agent Actor Critic
Shuo Liu, Tianle Chen, Ryan Amiri, Christopher Amato
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 a centralized critic instead of per-agent decentralized critics (ablation) | CoLLM · CoopHE | pass rate (%): 59.1 -> 75.2 | Improved |
| use a centralized critic instead of per-agent decentralized critics (ablation) | CoLLM · StrBuild (n=2) | IoU (%): 44.6 -> 68.5 | Improved |
| use a centralized critic instead of per-agent decentralized critics (ablation) | CoLLM · HouseBuild (n=2) | HP (%): 43.8 -> 86.4 | Improved |
| replace MA-REINFORCE (MAGRPO) with centralized-critic actor-critic | CoLLM-CC · CoopHE | pass rate (%): 74.3 -> 75.2 | No clear effect |
| replace MA-REINFORCE (MAGRPO) with centralized-critic actor-critic | CoLLM-CC · StrBuild (n=2) | IoU (%): 50.6 -> 68.5 | Improved |
| replace MA-REINFORCE (MAGRPO) with centralized-critic actor-critic | CoLLM-CC · TLDR | score (%): 93.5 -> 95.2 | No clear effect |
| fine-tune agents with MARL instead of prompt-only multi-agent | MAGRPO vs Parallel · TLDR | score (%): 22.9 -> 93.5 | Improved |
| fine-tune single model with GRPO | Qwen model · CoopHE | pass rate (%): 56.3 -> 61.8 | Improved |
| add a critic to single-model GRPO (AC) (ablation) | Single model · CoopHE | pass rate (%): 61.8 -> 62.5 | No clear effect |
| increase agents from 2 to 3 (ablation) | CoLLM-CC · StrBuild | IoU (%): 68.5 -> 75.4 | Improved |
| increase agents from 2 to 3 (ablation) | MAGRPO · StrBuild | IoU (%): 50.6 -> 38.1 | Hurt |
| increase agents from 2 to 3 (ablation) | CoLLM-DC · HouseBuild | IoU (%): 46.8 -> 43.7 | Hurt |
Closest papers by meaning
- Dr. MAS: Stable Reinforcement Learning for Multi-Agent LLM Systems (NeurIPS 2026 · similarity 0.91)
- ACC-Collab: An Actor-Critic Approach to Multi-Agent LLM Collaboration (ICLR 2025 · similarity 0.90)
- Stronger-MAS: Multi-Agent Reinforcement Learning for Collaborative LLMs (ICLR 2026 · similarity 0.90)
- Finite-Time Global Optimality Convergence in Deep Neural Actor-Critic Methods for Decentralized Multi-Agent Reinforcement Learning (ICML 2025 · similarity 0.89)
- Coevolving with the Other You: Fine-Tuning LLM with Sequential Cooperative Multi-Agent Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Investigating Relational State Abstraction in Collaborative MARL (AAAI 2025 · 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.
