On the Development of Multi Agent System for Large Urban Area Evacuation with Autonomous Navigation

Abstract

With the aim of estimating the effectiveness of the measures taken to smoothen and speedup the evacuation process of a large urban area, in time critical events like tsunami, a multi agent based mass evacuation simulation software is being developed. Considering the fact that it involves large number of human casualties, moderately complex agents in 2D grid environments are implemented. A vision based autonomous navigation algorithm, which enables the agents to move through an urban environment and reach a far visible destination, is implemented. The navigation algorithm is verified comparing the simulated evacuation time and the paths taken by individual agents with those of theoretical. We adopted optimal reciprocal collision avoidance algorithm for collision avoidance and validated by comparing with filed observations reported in literature. A parallel computing extension is developed for studying mass evacuation of large areas; simulating millions of agents with vision based navigation and collision avoidance is computationally intensive. A high strong scalability is attained with two millions of agents, which indicates potential to simulate mass evacuation involving many millions of people.

Agent Based Modeling and Simulation of Tsunami Triggered Mass Evacuation Considering Changes of Environment Due to Earthquake and Inundation

“…5. In order to make the agents use the high resolution grid models of complex urban environments, vision functionality is implemented 1) . Agents vision scans the environment like a radar, in high resolution, and deduces the available ways out in their visible surrounding, see Fig.…”

“…All the agent types, except the visitor, use the topological graph to find paths to a safe destination. The visitor agents do not have access to the graph, and either follow other moving agents in their vicinity or autonomously navigate towards far visible destinations deducing the path based on what they see in their visible surrounding 1) . For the purpose of visitor agent's following, it is assumed that, when choosing a person to be followed, the visitors do not know if the person they are following know their destination, i.e.…”

“…With the aim of exploring the effectiveness of different strategies to make the evacuation process, in large urban areas, smooth and fast, we are developing an agent based evacuation simulation code 1) . In order to overcome the limitations of existing mass evacuation simulators, a detailed model of the environment and agents with functionalities to interact with it and other agents are included.…”