A Novel Intelligent Traffic Recovery Model for Emergency Vehicles Based on Context-aware Reinforcement Learning

Abstract

Management of traffic emergencies has become very popular in recent years. However,

timely response to emergencies and recovering from an emergency is an important prob- lem in itself. The strategies in the current studies merely suggest that after an emergency

vehicle passes, the state should iterate to the next phase. Therefore, this paper proposes a novel approach for recovering from an emergency situation at an intersection based on real scenarios. The proposed method is a combination of context-aware and Reinforcement Learning (RL) models that predicts better alternatives for different states rather than just iterating to the next phase. In this regard, a new algorithm, named Interrupt Algorithm,

is proposed to predict proper actions for recovering the emergency situation. This algo- rithm uses a Q-learning-based model that learns from traffic context for an emergency sit- uation and chooses viable action from an action set. The recovery actions are categorized as

max, min, and avg, respectively. Test results show that our proposed model outperforms

traffic flow over than standard single choice recovering action-based approach by approx- imately 80%. Based on this, it may be more beneficial to choose different actions and there- fore, proposed algorithm with the help of RL presents a more dynamic emergency recovery

model.