Personalized lane change decision algorithm using deep reinforcement learning approach

To develop driving automation technologies for humans, a human-centered methodology should be adopted safety and satisfactory user experience. Automated lane change decision in dense highway traffic is challenging, especially when considering different driver preferences. This paper proposes personalized algorithm based on deep reinforcement learning. Firstly, experiments are carried out moving-base simulator. Based the analysis of experiment data, three personalization indicators selected to describe preferences lane-change decisions. Then, learning (RL) approach applied design human-like agents automated decisions capture preferences, with refined rewards using indicators. Finally, trained RL benchmark tested two-lane scenario. Results show that proposed can achieve higher consistency than comparison algorithm.