Before Humans Join the Team: Diagnosing Coordination Failures in Healthcare Robot Team Simulation
Summary: arXiv:2508.04691v2 Announce Type: replace-cross
Abstract
As humans move toward collaborating with coordinated robot teams, understanding how these teams coordinate and fail is essential for building trust and ensuring safety. However, exposing human collaborators to coordination failures during early-stage development is costly and risky, particularly in high-stakes domains such as healthcare.
Introduction
In recent years, the integration of robotic systems in healthcare settings has garnered significant attention. While the potential benefits of robotic assistance are promising, the complexities of team coordination pose challenges that must be addressed before human operators can safely engage with these automated systems. The study presented in this article adopts an innovative agent-simulation approach to diagnose and understand coordination failures in robotic teams before human involvement.
Methodology
In our research, we instantiate all team roles, including the supervisory manager, as large language model (LLM) agents. This allows us to simulate interactions and diagnose potential coordination failures prior to human integration. We constructed a controllable healthcare scenario and conducted two studies with varying hierarchical configurations to analyze coordination behaviors and failure patterns.
Findings
Our findings highlight several key aspects regarding coordination failures in robotic teams:
- Team Structure: We discovered that team structure is the primary bottleneck for effective coordination, overshadowing contextual knowledge or model capability.
- Reasoning Autonomy: The study revealed a significant tension between the autonomy of reasoning among team members and the overall stability of the system.
- Simulation Benefits: By surfacing these failures through simulation, we establish a foundational understanding that prepares the ground for safe human integration.
Implications
The implications of our findings are profound, particularly for the design of resilient robot teams. Our research informs several critical areas:
- Process-Level Evaluation: Enhanced strategies for evaluating the coordination processes within robotic teams.
- Transparent Coordination Protocols: Development of clearer protocols that promote effective communication and coordination among robot team members.
- Structured Human Integration: Establishing a structured approach for integrating human operators into robotic teams, ensuring safety and trust.
Conclusion
As we advance toward a future where humans and robots collaborate in healthcare, understanding the coordination dynamics of robotic teams is crucial. Our work provides a framework for diagnosing coordination failures and prepares the path for the safe integration of human collaborators. The supplementary materials, including codes, task agent setup, trace outputs, and annotated examples of coordination failures and reasoning behaviors, can be accessed at https://byc-sophie.github.io/mas-to-mars/.