Weighted Affordance-based Agent Modeling and Simulation in Emergency Evacuation

(submitted to Safety Science)

Moise Busogi, Dongmin Shin, Hokyoung Ryu, Yeonggwang Oh & Namhun Kim

hdr_img

Abstract: This paper presents an agent-based human behavioral modeling framework to analyze probable human actions, in emergency situations, considering both physical and psychological dimensions, in emergency situations. Human's prospective controls suggest that the environment can offer certain physical and psychological conditions to opt for a finite number of feasible human actions that lead to desired system states. A set of possible human actions is then generated and updated from the affordance-effectivity duals in a spatial-temporal dimension. In this paper, a reward and cost-based dynamic affordance-based agent model is built upon physical and psychological constraints that are inserted for the agents' decision-making processes. The model employs Markov Decision Process (MDP), and NASA-TLX (Task Load Index) is used as cost and reward estimates. The action selection process of human agents, i.e., triggering of state transitions, is stochastically modeled in accordance with the action-state cost (load) values. A series of affordance-based numerical values are calculated for predicting prospective actions in the system. Finally, an evacuation simulation example based on the proposed model is illustrated to verify the proposed human behavioral modeling framework.

The model proposed in this study is an agent-based modeling of human-involving complex and dynamic systems. The proposed framework covers dynamic decision-making processes with considering affordance-embedded MDP (Markov Decision Process). We adopt the probability-based action selection model to present the uncertainty of human actions within dynamic environments.
Link to read more, please contact me via email.