GlobeDiff: State Diffusion Process for Partial Observability in Multi-Agent Systems

Yiqin Yang; Xu Yang; Yuhua Jiang; Ni Mu; Hao Hu; Runpeng Xie; Ziyou Zhang; Siyuan Li

cs.AIFeb 17, 2026

GlobeDiff: State Diffusion Process for Partial Observability in Multi-Agent Systems

Authors: Yiqin Yang, Xu Yang, Yuhua Jiang, Ni Mu, Hao Hu, Runpeng Xie, Ziyou Zhang, Siyuan Li,

Organizations: The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences · Tsinghua University · Moonshot AI · Faculty of Computing, Harbin Institute of Technology ·

Abstract

In the realm of multi-agent systems, the challenge of partial observability is a critical barrier to effective coordination and decision-making. Existing approaches, such as belief state estimation and inter-agent communication, often fall short. Belief-based methods are limited by their focus on past experiences without fully leveraging global information, while communication methods often lack a robust model to effectively utilize the auxiliary information they provide. To solve this issue, we propose Global State Diffusion Algorithm~(GlobeDiff) to infer the global state based on the local observations. By formulating the state inference process as a multi-modal diffusion process, GlobeDiff overcomes ambiguities in state estimation while simultaneously inferring the global state with high fidelity. We prove that the estimation error of GlobeDiff under both unimodal and multi-modal distributions can be bounded. Extensive experimental results demonstrate that GlobeDiff achieves superior performance and is capable of accurately inferring the global state.

Figures & tables

Figure 1: The overall framework of GlobeDiff. During the execution phase, we first construct auxiliary local observations $x$ and then infer the global state $\hat{s}$ using GlobeDiff. Agents make decisions based on the inferred global state $\hat{s}$ .

Figure 2: The training process of Globediff is divided into two parts: minimizing the difference between the prior network $p_{\phi}$ and the posterior network $q_{\psi}$ , and then training the diffusion model based on the forward and backward process.

Panel of Figure 3 — Figure 3: Comparison results with global state inference baselines in SMAC-v1 (PO) tasks with win rate over three random seeds.

Panel of Figure 4 — Figure 4: Comparison results with global state inference baselines in SMAC-v2 (PO) tasks with win rate over three random seeds.

Figure 5: Visualization of global states generated by GlobeDiff, VAE and MLP. The first plot displays true states and subsequent plots show inferred states per agent. White points denote individual states with polygons highlighting local neighborhoods. Gradient shading (light green to purple) indicates training progression. The similarity between the polygon structures of the inferred and true states reflects the predicted quality.

Panel of Figure 6 — Figure 6: Comparison results with generative model baselines in SMAC-v1 (PO) tasks with win rate over three random seeds.

Panel of Figure 7 — Figure 7: Comparison results with generative model baselines in SMAC-v2 (PO) tasks with win rate over three random seeds.

Figure 8: Win rate with various sight range in original SMAC tasks.

Figure 9: Ablation of prior network.

Panel of Figure 10 — Figure 10: Ablation study for diffusion step and residual blocks on the zerg 5v5 task.

Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 11: Experiments of the qualitative analysis.

	Vanilla MAPPO	Vanilla MAPPO (Large)	GlobeDiff
zerg 5v5	0.22 $\pm$ 0.01	0.23 $\pm$ 0.00	0.33 $\pm$ 0.02
protoss 5v5	0.21 $\pm$ 0.02	0.24 $\pm$ 0.01	0.38 $\pm$ 0.01
terran 5v5	0.16 $\pm$ 0.01	0.17 $\pm$ 0.00	0.24 $\pm$ 0.01
zerg 10v10	0.13 $\pm$ 0.01	0.15 $\pm$ 0.01	0.25 $\pm$ 0.01
zerg 10v11	0.06 $\pm$ 0.01	0.07 $\pm$ 0.01	0.12 $\pm$ 0.01
terran 10v11	0.02 $\pm$ 0.00	0.03 $\pm$ 0.00	0.07 $\pm$ 0.01

Appendix

Table 1: Comparison results between Vanilla MAPPO, Vanilla MAPPO (Large), and GlobeDiff

Figure 12: U-Net architecture in Diffusion model.

Hyper-parameters	Value	Hyper-parameters	Value	Hyper-parameters	Value
Critic Learning Rate	5e-4	Actor Learning Rate	5e-4	Use GAE	True
Gain	0.01	Optim	1e-5	Batch Size	64
Training Threads	4	Num Mini-Batch	1	Rollout Threads	8
Entropy Coef	0.01	Max Grad Norm	10	Episode Length	400
Optimizer	Adam	Hidden Layer Dim	64	GAE $\lambda$	0.95
Activation Function	Relu	PPO Epoch	15	$\gamma$	0.99

Appendix

Table 2: Hyper-parameters for MAPPO.

Prior Network $p_{\phi}$	Value	Posterior Network $q_{\psi}$	Value
$z$ dim	16	$z$ dim	16
Hidden Layer Dim	1024	Hidden Layer Dim	1024
Hidden Layer Num	3	Hidden Layer Num	3
Activation Function	Relu	Activation Function	Relu
Learning Rate	2e-4	Learning Rate	2e-4
Weight Decay	1e-4	Weight Decay	1e-4

Appendix

Table 3: Hyper-parameters for prior network

p_{\phi}

and posterior network

q_{\psi}

GlobeDiff	Value	VAE	Value	MLP	Value
Down-sampling Factor	8	Latent Dim	256	Hidden Layer Dim	1024
Diffusion Steps	5	Hidden Layer Dim	1024	Hidden Layer Num	4
Residual Blocks	2	Activation Function	Relu	Activation Function	Relu
Learning Rate	2e-4	Learning Rate	1e-4	Learning Rate	3e-4
Batch Size	32	Batch Size	32	Batch Size	32

Appendix

Table 4: Hyper-parameters for Generative methods.

Multi-Agent Systems

Multi-Agent Systems studies how autonomous agents coordinate, compete, communicate, and learn in shared or interdependent tasks.

See all 677 papers

Global Coherence: When Every Agent Is Right and the Team Is Still Wrong - A Local-to-Global Semantic Foundation for Multi-Agent Collaboration
Oct 1, 2026Xin HengMulti-Agent Collaboration Sheaf Theory
Tote AI
LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning
May 18, 2026Sangjun Bae, Yisak Park, Sanghyeon Lee +1LLM-Guided RL Multi-Agent Systems
Graduate School of Artificial Intelligence, UNIST, Ulsan, South Korea.
CoDiMAD: Diffusion-Based Privileged Distillation for Communication-Free Multi-Robot Coordination
Jul 10, 2026Jiyue Tao, Shunheng Xin, Tongsheng Shen +2Multi-Agent Coordination Multi-Robot Coordination
Robotics and Control Laboratory, School of Advanced Manufacturing and Robotics, and the State Key Laboratory of Turbulence and Complex Systems, Peking University, Beijing, 100871, China · National Innovation Institute of Defense Technology, Beijing 100071, China

Oct 1, 2026J/K move · Enter open · S save

Global Coherence: When Every Agent Is Right and the Team Is Still Wrong - A Local-to-Global Semantic Foundation for Multi-Agent Collaboration

Xin Heng

Tote AI

Multi-Agent Collaboration Sheaf Theory Multi-Agent Systems

Open paper AskPDF arXiv

AI agents can each make locally valid decisions yet jointly produce an invalid result. We call this the global coherence problem: a failure of shared state, not merely of intelligence. Our Observation-Aliasing Impossibility Theorem gives the exact boundary. A policy can guarantee a valid action exactly when all worlds producing the same observation share an admissible action. If k indistinguishable worlds require pairwise-disjoint actions, the best randomized worst-case success is 1/k; more reasoning, roles, messages, or samples cannot recover the missing distinction. A stronger model can reason better within its context, but it cannot see beyond it. The smallest sufficient context is NP-hard to find, yet counterexample-guided refinement finds a sufficient one and recertifies it as rules change. We give local-to-global runtime semantics X = (H, C, G, F; D): topology H records overlapping scopes; category C governs state-changing actions; groupoid G keeps reversible translations; sheaf F tests whether local views glue into one world; and history D keeps only what alters legal futures. Each preserves the needed information as state is split, translated, or changed. Models propose; the harness owns shared state and governs commit. Ten studies test the boundary. On a revision benchmark, a frontier model scores 40/40 when the deciding event is visible; hidden, every arm is consistent with chance (1/3); one restored fact returns 40/40. On TeamBench, ordinary teams exceed a shared budget in 5/5 runs, a visible live count leaves 4/5, and commit enforcement leaves 0/5. In tau2-bench Telecom, current-state checks score 0.07 after silent reverts; the harness scores 1.00. Refinement keeps a sufficient context through rule changes at 28% of the all-field cost. Solvers that already own the state tie it, as predicted. Local intelligence cannot substitute for missing global state.

Abstract

Multi-Agent Systems

Explore similar work

Global Coherence: When Every Agent Is Right and the Team Is Still Wrong - A Local-to-Global Semantic Foundation for Multi-Agent Collaboration

LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning

CoDiMAD: Diffusion-Based Privileged Distillation for Communication-Free Multi-Robot Coordination