Real‐time Deep Reinforcement Learning for Evacuation under Emergencies

Abstract

Aviation emergencies pose high risks to humans, when it occurs, it is imperative to evacuate humans to safe places in an efficient manner. Under the high level of time pressure, decision‐makers are facing great challenges in developing an optimal evacuation quickly, especially when the threats involved are not static and the environment is unfamiliar. We plan to integrate Asynchronous Advantage Actor Critic (A3C) algorithm with the velocity obstacle (VO) models to optimize evacuation in an airport environment under emergencies. A multi‐agent collaborative evacuation modeling framework for a complex environment with moving threats will be developed to provide adaptive continuous decision‐aid to each agent, in accordance to the changing environments.

CATM Research Affiliates:
Dahai Liu (ERAU: Lead)
Sirish Namilae (ERAU)