Researchers propose GPEvac, a graph neural network framework for real-time adaptive evacuation routing during active shooter events

Researchers from arXiv:2609.16163 propose GPEvac, a graph neural network–based proximal policy optimization (PPO) framework for computing adaptive evacuation routes during active shooter events. The system is designed to minimize threat exposure while accounting for adversarial uncertainty and crowding dynamics. Unlike prior methods constrained to specific building layouts or computationally intractable at scale, GPEvac uses a single learned policy that generalizes across diverse architectural topologies and sizes.

The framework introduces an edge-first sequential message-passing scheme with a learnable virtual global node to capture both local and long-distance dependencies in evacuation environments. These graph embeddings feed into a permutation-invariant scoring mechanism, enabling the policy to operate across varied layouts without retraining.

In extensive simulations, GPEvac outperformed intelligent baselines across distinct architectural layouts, significantly reducing total threat exposure. The authors report that the system computes global evacuation routes in 14.73 milliseconds on local CPU hardware, a latency compatible with integration into live surveillance systems.

Beyond active shooter scenarios, the authors note that the methodologies developed are transferable to other graph-structured decision-making domains, including critical infrastructure, intelligent transportation systems, and adaptive sensor networks.