Environment-sensitive crowd behavior modeling method based on reinforcement learning.

Autor:	Pang, Chen, Lyu, Lei, Zhou, Qinglin, Zhou, Limei
Předmět:	COLLECTIVE behavior REINFORCEMENT learning CIVILIAN evacuation LEARNING strategies ACCESS to information
Zdroj:	Applied Intelligence; Aug2023, Vol. 53 Issue 16, p19356-19371, 16p
Abstrakt:	Most existing crowd evacuation methods focus on internal factors and do not consider the influence of the external environment factors, producing unrealistic global behavior occurs when individuals are moving through the crowded space. As an essential part of a building, safety indication signs (SISs) are a form of environmental information and perceptual access that play an important role in promoting wayfinding by virtue of their guiding role in movement direction and route selection by providing guidance, warning and mandatory message to people. In this paper, we propose an innovative crowd simulation method guided by SISs with reinforcement learning strategy for use in emergencies. To illustratethe guiding function of the SISs, we establish a guidance field for each SIS and add the attractive force to the guidance field. Besides, we formulate the multi-agent (crowd) navigation problem as an action-selection problem and design a novel reinforcement learning strategy for driving individuals to accomplish collision-free movement more efficiently. Particularly, we define a state transition strategy between the SIS area and the non-SIS area to achieve continuity of guidance. We use extensive simulations to highlight the potential of our method in different scenarios and evaluate the results in terms of evacuation efficiency and the reasonableness of SIS placement. In practice, our system runs at interactive rates and can solve complex planning problems involving one or more groups. [ABSTRACT FROM AUTHOR]
Databáze:	Complementary Index
Externí odkaz:	Zobrazit plný text záznamu Full text from SpringerLink