📄 中文摘要
AgentComm-Bench是一个新的基准测试套件，旨在在六种现实网络干扰下对多智能体体化AI系统进行压力测试。研究发现，面对现实世界中不完美的通信网络，最先进的合作体智能系统在导航和感知F1评分上表现出超过96%和85%的性能下降。这一发现揭示了实验室评估与可部署系统之间的重大差距，强调了在实际应用中需要解决的关键问题。
AgentComm-Bench揭示了合作体智能在现实网络条件下的灾难性失效模式
📄 English Summary
AgentComm-Bench Exposes Catastrophic Failure Modes in Cooperative Embodied AI Under Real-World Network Conditions
AgentComm-Bench is a new benchmark suite designed to stress-test multi-agent embodied AI systems under six real-world network impairments. The research reveals that state-of-the-art cooperative embodied AI systems, which are intended for use in robots, drones, and autonomous vehicles, exhibit catastrophic brittleness when confronted with the imperfect communication networks of the real world. Performance drops of over 96% in navigation and 85% in perception F1 scores were observed, highlighting a critical gap between laboratory evaluations and deployable systems. This underscores the urgent need to address these vulnerabilities for practical applications.