MACRO-LLM: LLM-Empowered Multi-Agent Collaborative Reasoning under Spatiotemporal Partial Observability

TL;DR

MACRO-LLM tackles coordination of distributed LLM agents under spatiotemporal partial observability by decomposing reasoning into three modules: CoProposer for rollout-verified proposals, Negotiator with mean-field aggregation for spatial alignment, and Introspector with semantic gradient descent for temporal adaptation. The framework enables decentralized, zero-shot coordination without central controllers, validated on Cooperative Adaptive Cruise Control and Pandemic Control tasks where it achieves superior coordination, scalability, and robustness across topologies and backbone models. Key innovations include rollout-based verification to navigate future uncertainty, mean-field statistical features for scalable neighborhood communication, and semantic-gradient-driven updates to adapt strategies over time. The results demonstrate practical impact for large-scale, real-world multi-agent systems requiring long-horizon planning under partial observability, with clear pathways for extension to additional domains and more efficient prompts or local model backends.

Abstract

Large Language Model (LLM) agents deployed in complex real-world scenarios typically operate as spatially distributed entities. However, this physical dispersion constrains agents to limited local perception and finite temporal horizons. We characterize this bottleneck as spatiotemporal partial observability. Given such fragmented awareness, distributed agents struggle to coordinate efficiently. To bridge this gap, we introduce MACRO-LLM, LLM-empowered multi-agent collaborative reasoning under spatiotemporal partial observability. The architecture addresses spatiotemporal constraints via three modules: (1) the CoProposer mitigates temporal uncertainty by verifying candidate actions via predictive rollouts; (2) the Negotiator overcomes spatial myopia by resolving conflicts through mean-field statistical aggregation; and (3) the Introspector ensures continuous adaptation by analyzing historical experience to refine strategies via semantic gradient descent. Extensive evaluations on two complex long-horizon tasks, cooperative adaptive cruise control and pandemic control, demonstrate that our framework effectively mitigates spatiotemporal partial observability through spatial and temporal strategies, enabling robust coordination.

Figures (12)

• Figure 1: Illustration of spatiotemporal partial observability and the proposed collaborative reasoning framework. (a) Agents are constrained by limited local perception (spatial) and future uncertainty coupled with finite context windows (temporal). (b) Agents mitigate these limitations by exchanging proposals to establish collaborative framework via neighborhood negotiation.
• Figure 2: Architecture of MACRO-LLM for agent $n$. The framework comprises three synergistic modules: (1) the CoProposer generates proposals; (2) the Negotiator handles conflict resolution and spatial strategy updates; and (3) the Introspector performs strategy refinement. The CoProposer and Negotiator form a negotiation loop to process observations at time $t$ and output coordinated actions, leading to the next observation at $t+1$.
• Figure 3: Scalability analysis of MACRO-LLM on the CACC task for number of agents $N \in \{8, 16, 24, 32\}$.
• Figure 4: Ablation of MACRO-LLM—impact of each module on CACC performance.
• Figure 5: Ablation of MACRO-LLM—impact of each module on PC performance