SyncPlan: Long-Horizon LLM Coordination with Explicit Synchronization and Adaptive Correction

2026-08-02

AI Summary

This work addresses the challenge of balancing efficiency and adaptability in large language model (LLM)-driven multi-agent coordination within dynamic environments. To this end, we propose SyncPlan, a framework that generates action chains for all agents through a single invocation of a centralized LLM, augmented with explicit synchronization primitives, a lightweight plan staleness detector, and an adaptive replanning mechanism. This integration enables highly efficient and robust long-horizon collaboration at minimal computational overhead. SyncPlan is the first to jointly incorporate explicit synchronization, deadlock detection, and planning optimization, leveraging both supervised fine-tuning and planning-oriented reinforcement learning. Evaluated on the Overcooked and Honor of Kings benchmarks, SyncPlan achieves state-of-the-art task success rates while consuming less than 0.05% of the runtime required by existing LLM-based coordination approaches.

Abstract

LLM-based multi-agent coordination faces a fundamental trade-off between efficiency and adaptivity in dynamic environments. Existing approaches typically rely on repeated LLM invocations or multi-round communication to adapt decisions during execution, introducing substantial latency and making coordination vulnerable to asynchronous progress and environmental changes. Conversely, one-shot planning reduces coordination overhead but produces open-loop plans that can quickly become stale or fail when actions depend on other agents and the environment. We introduce SyncPlan, a plan-execute-correct framework for long-horizon coordination through explicit synchronization and adaptive correction. Given the state and team-level task, a centralized LLM coordinator generates per-agent action chains in a single planning call. During execution, explicit wait primitives and deadlock detection enforce inter-agent and agent-environment dependencies, while a lightweight Plan Staleness Detector continuously assesses the remaining plan and triggers replanning when environmental changes invalidate its assumptions. We further optimize the coordinator through SFT and planning-oriented RL with dense task progress and outcome-level execution feedback. Experiments on the public Overcooked benchmark and the complex Honor of Kings environment show that SyncPlan achieves state-of-the-art task success rates while using less than 0.05% of the wall-clock runtime compared with existing LLM-based coordinators. Code and datasets will be made publicly available.

Problem

• LLM-based multi-agent coordination
• efficiency-adaptivity trade-off
• dynamic environments
• open-loop planning
• asynchronous progress

Innovation

• SyncPlan
• LLM-based coordination
• explicit synchronization
• adaptive replanning
• plan staleness detection