Routing algorithms as tools for integrating social distancing with emergency evacuation.

Area of Science:

• Operations Research
• Disaster Management
• Artificial Intelligence

Background:

• The COVID-19 pandemic highlighted the necessity of social distancing, even during critical events like pre-hurricane evacuations.
• Integrating social distancing into evacuation logistics presents significant operational challenges.

Purpose of the Study:

• To explore the implications of incorporating social distancing into evacuation operations.
• To compare the efficiency of Deep Reinforcement Learning (DRL) against traditional algorithms for social-distanced evacuations.

Main Methods:

• The evacuation process was modeled as a Capacitated Vehicle Routing Problem (CVRP).
• A Deep Neural Network (DNN)-based solution (Deep Reinforcement Learning) was employed.
• A non-DNN solution, the Sweep Algorithm, was used for comparison.

Main Results:

• Deep Reinforcement Learning demonstrated more efficient routing compared to the Sweep Algorithm.
• However, the time saved by DRL routing was insufficient to compensate for the additional time required for social distancing.
• The efficiency advantage of DRL diminished as emergency vehicle capacity neared typical household sizes.

Conclusions:

• While DRL offers routing improvements, it does not fully resolve the time constraints imposed by social distancing in evacuations.
• The effectiveness of DRL in such scenarios is highly dependent on vehicle capacity and the degree of social distancing required.