Open and real-world human-AI coordination by heterogeneous training with communication

Human-AI coordination aims to develop AI agents capable of effectively coordinating with human partners. Achieving satisfying performance of AI agents in open and real-world environments poses a long-standing challenge. To facilitate the practical deployment and application of human-AI coordination in open and real-world environments, Yang Yu from LAMDA Group of Nanjing University published their new research on 15 Apr 2025 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.

The team propose the problem formulation and a first benchmark for open and real-world human-AI coordination (ORC). The benchmark, called ORCBench, includes popular environments that are commonly used in recent works. It not only considers the open challenge, i.e., different unseen partners with different unknown levels, but also the real challenge, i.e., agents and partners are heterogeneous, from the observation to the capability.

Furthermore, they introduce a framework known as Heterogeneous training with Communication (HeteC) for ORC. HeteC builds upon a heterogeneous training framework and enhances partner population diversity by using mixed partner training and frozen historical partners. Additionally, HeteC incorporates a communication module that enables human partners to communicate with AI agents, mitigating the adverse effects of partially observable environments.

To validate the effectiveness of HeteC, they conducted experiments on ORCBench and compared it with existing methods. Firstly, they demonstrate that the ORC environment poses challenges for the state-of-the-art HAC methods. Next, they evaluate the effectiveness of the proposed method on different layouts and with various masks (i.e., different limited observation spaces of agents), including experiments with real-human participants. Their contribution serves as an initial but important step towards addressing the challenges of ORC.

DOI: 10.1007/s11704-024-3797-6