ReCollab: Retrieval-Augmented LLMs for Cooperative Ad-hoc Teammate Modeling

TL;DR

This paper presents ReCollab, a retrieval-augmented language model framework for better modeling and adapting to unseen teammates in cooperative ad-hoc teamwork scenarios, demonstrating improved accuracy and performance.

Contribution

It introduces ReCollab, a novel retrieval-augmented LLM approach for teammate modeling that enhances inference stability and adaptation in ad-hoc teamwork environments.

Findings

ReCollab effectively distinguishes teammate types in Overcooked.

ReCollab improves adaptation performance across different layouts.

The framework achieves Pareto-optimal trade-offs between accuracy and return.

Abstract

Ad-hoc teamwork (AHT) requires agents to infer the behavior of previously unseen teammates and adapt their policy accordingly. Conventional approaches often rely on fixed probabilistic models or classifiers, which can be brittle under partial observability and limited interaction. Large language models (LLMs) offer a flexible alternative: by mapping short behavioral traces into high-level hypotheses, they can serve as world models over teammate behavior. We introduce \Collab, a language-based framework that classifies partner types using a behavior rubric derived from trajectory features, and extend it to \ReCollab, which incorporates retrieval-augmented generation (RAG) to stabilize inference with exemplar trajectories. In the cooperative Overcooked environment, \Collab effectively distinguishes teammate types, while \ReCollab consistently improves adaptation across layouts, achieving Pareto-optimal trade-offs between classification accuracy and episodic return. These findings demonstrate the potential of LLMs as behavioral world models for AHT and highlight the importance of retrieval grounding in challenging coordination settings.