AI Paper Atlas

Research map / Multi-agent systems and games / LLM agents

Distilling Task-Level Coordination Policies for Generalizable Multi-Agent Cooperation

Zimo Zhai, Manjie Xu, Wei Liang

ICML 2026 · Multi-agent systems and games › LLM agents · approach: Robust and efficient agent execution

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes SynCoord, a self-supervised pipeline that distills LLM-derived task-level coordination policies into smaller models for generalizable multi-agent cooperation.

Problem

Directly deploying LLMs as high-level coordinators for multi-agent systems is hard because effective policies often fail to emerge reliably at the low-level control stage, and inference costs limit scalability.

Approach

Self-supervised synthetic coordination distillation (SynCoord) that transfers task-level coordination policies into a model for multi-agent cooperation, avoiding costly per-step LLM inference. Details beyond the abstract are not available.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add the single-agent bootstrap phase (ablation)SynCoord · Overcooked-AIsuccess rate: 0.65 -> 0.95Improved
add the single-agent bootstrap phase (ablation)SynCoord · Overcooked-AIteacher data accuracy: 0.90 -> 0.97Improved
use 235B teacher instead of 30B teacher (ablation)SynCoord · Overcooked-AIsuccess rate: 0.92 -> 0.95Improved
use 235B teacher instead of 8B teacher (ablation)SynCoord · Overcooked-AIsuccess rate / data accuracy: SR 0.80 -> 0.95; Data Acc 0.69 -> 0.97Improved
use 7B intuitive agent instead of 1.5B (ablation)SynCoord · Overcooked-AIsuccess rate: 0.00 -> 0.95Improved
distill task-level coordination into a 7B model instead of prompting the 235B thinking agentSynCoord · Overcooked-AIdecision latency: 15609 ms -> 350 msImproved
distilled SynCoord versus prompted LLMs on 5x13 layoutsSynCoord · Overcooked-AI 5x13avg success timesteps: 39.97 vs 44.24 (Gemini-3-flash), 46.88 (Qwen3-235B), 48.25 (GPT-5.2)Improved
expand training layouts from 1x1 to 1x13 for RL baselinesMAPPO · Overcooked-AIsuccess rate: 1.00 -> 0.21Hurt
unseen versus seen recipesSynCoord · Overcooked-AI 17 unseen recipessuccess rate: 0.98 -> 0.97No clear effect
zero-shot transfer from 1-2 agents to larger LBF layout with 100 stepsSynCoord · LBF 15x15-3p-5fsuccess rate: 0.00 -> 0.80Improved
increase team size to three agentsSynCoord · Overcooked-AI random layoutssuccess rate: near-perfect -> 82.8%Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy