Grounded Joint-Attention Other-Play for Zero-Shot Coordination

2026-10-05

AI Summary

This study addresses the fragility of multi-agent zero-shot coordination, where reliance on arbitrary conventions formed during training often leads to poor collaboration with unseen partners. To overcome this limitation, this work proposes the MATE framework, which draws inspiration from human joint attention. By integrating multi-agent reinforcement learning with visual attention mechanisms, MATE aligns environment-grounding signals to guide agents toward focusing on salient objects in the scene, thereby achieving environment-anchored coordination. This approach proactively establishes a collaborative paradigm grounded in environmental cues, extending beyond traditional symmetry-breaking methods that merely prevent convention fixation. Experimental evaluations demonstrate that MATE significantly enhances coordination capabilities with novel, unseen partners on benchmarks such as Card Alignment, consistently outperforming existing state-of-the-art methods.

Abstract

Joint attention - the human ability to share a common visual or cognitive focus with others - enables a meeting of minds that lets us coordinate even with unfamiliar partners. In this work we investigate whether equipping AI agents with a similar mechanism can enable such zero-shot coordination. We introduce Mutual Attention for zero-shot TEaming (MATE): a novel multi-agent reinforcement learning method inspired by human joint attention. MATE encourages agents to coordinate their actions by aligning their visual attention on scene-salient objects during the interaction rather than relying on arbitrary partner-dependent conventions established during training. Unlike symmetry-breaking approaches that merely prevent brittle conventions from emerging, MATE actively promotes coordination through an environment-grounded signal that is naturally shared across partners. We evaluate MATE on three benchmarks: our Card Alignment Game, designed to isolate brittle convention formation, and the more challenging Level-Based Foraging and OvercookedV2 benchmarks. Our experiments consistently show that a joint-attention-inspired signal improves coordination with unknown partners, underlining MATE's potential as a general coordination mechanism that complements and surpasses symmetry-breaking approaches.

Problem

• zero-shot coordination
• joint attention
• multi-agent reinforcement learning
• brittle conventions

Innovation

• Joint Attention
• Zero-Shot Coordination
• Multi-Agent Reinforcement Learning
• Mutual Attention
• Symmetry Breaking