Research map / Multi-agent systems and games / Cooperative multi-agent RL
GlobeDiff: State Diffusion Process for Partial Observability in Multi-Agent System
Yiqin Yang, Xu Yang, Yuhua Jiang, Ni Mu, Hao Hu, Runpeng Xie, Ziyou Zhang, Siyuan Li, Yuan-Hua Ni, Qianchuan Zhao, Bo XU
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 |
|---|---|---|---|
| add GlobeDiff global state diffusion to MAPPO | MAPPO · SMAC-v2 (PO) zerg 5v5 | win rate: 0.22 -> 0.33 | Improved |
| add GlobeDiff to MAPPO | MAPPO · SMAC-v1 (PO) 6h vs 8z | win rate: 0.12 -> 0.47 | Improved |
| scale MAPPO networks to match GlobeDiff parameter count (ablation) | MAPPO · SMAC (PO), 9 maps | win rate: zerg 5v5 0.22 -> 0.23; MMM2 0.27 -> 0.01 | No clear effect |
| replace GlobeDiff with conditional VAE or MLP (ablation) | GlobeDiff · SMAC (PO) | win rate | Hurt |
| remove prior network and KL constraint (ablation) | GlobeDiff · SMAC (PO) various maps | win rate | Improved |
| increase diffusion steps K from 1 to 8 (ablation) | GlobeDiff · SMAC-v2 (PO) zerg 5v5 | state inference accuracy | Improved |
| increase number of U-Net residual blocks (ablation) | GlobeDiff · SMAC-v2 (PO) zerg 5v5 | win rate | No clear effect |
| feed joint observations of agents to policy (MAPPO Joint) (ablation) | MAPPO · SMAC-v2 (PO) | win rate | Mixed |
| narrow sight range from 9 to 3 in original SMAC (ablation) | MAPPO · SMAC-v2 | win rate: 0.03 drop | No clear effect |
Closest papers by meaning
- Learning POMDP World Models from Observations with Language-Model Priors (NeurIPS 2026 · similarity 0.86)
- Factored Online Planning in Many-Agent POMDPs (AAAI 2024 · similarity 0.85)
- Autoregressive Diffusion World Models for Off-Policy Evaluation of LLM Agents (NeurIPS 2026 · similarity 0.84)
- Revisiting Multi-Agent World Modeling from a Diffusion-Inspired Perspective (NeurIPS 2025 · similarity 0.84)
- ESCORT: Efficient Stein-variational and Sliced Consistency-Optimized Temporal Belief Representation for POMDPs (NeurIPS 2025 · similarity 0.84)
- Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy (NeurIPS 2024 · similarity 0.84)
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.
