Robust, privacy-aware communication protocols for multi-edge-device safety agents

Develop and rigorously evaluate inter-device communication protocols for agent-based edge video analysis systems deployed across multiple edge devices that exchange event metadata, confidence scores, and high-level alerts; ensure that these protocols are robust, privacy-preserving, and resilient to partial failures while maintaining synchronization, low latency, reliability, and security in real-world public safety deployments.

Background

The paper proposes a hybrid edge-based action detection system that combines skeleton-based motion analysis with vision-LLMs, with an intended agent-based architecture to coordinate perception and reasoning components. While raw video is processed and retained locally on each edge device to preserve privacy, practical deployments may involve multiple edge devices operating as coordinated agents.

In such distributed settings, the authors highlight that inter-device communication may involve exchanging event metadata, confidence scores, or high-level alerts to improve global situational awareness. However, this introduces dependencies related to synchronization, latency, reliability, and security. As explicitly stated, designing communication protocols that meet robustness, privacy, and partial-failure resilience requirements remains an open challenge, motivating this open problem.