Modelling herd behaviour in traffic jams using Markov chains-based reinforcement learning

Traffic congestion remains a persistent problem that compromises road safety. This phenomenon is often amplified by driver behaviours, particularly those characterised by the herd effect. This study aims to model the emergence and dynamics of the herd effect in traffic jams and to simulate a strategy for mitigating this behaviour among drivers. To achieve these objectives, reinforcement learning (RL) was employed within the frameworks of memoryless Markov chains and multi-phase Markov chains. The results demonstrate the effectiveness of Markov chains in accurately modelling the collective behaviour of specific drivers. Likewise, the simulations illustrate RL's capacity to regulate the herd effect and optimise individual decision-making during congestion. The findings suggest that traffic authorities may consider implementing RL-based strategies to mitigate herd behaviour, improve traffic flow, and enhance road safety.