A Personalized Ramp Merging Decision-Making Method for Autonomous Driving Based on Reverse Reinforcement Learning

In the ramp merging scenario, the merging vehicles need to make decisions during the interaction with high-speed vehicles on the main lane to achieve safe and reliable merging. The advanced driving assistance system can assist in decision-making during this process, providing reference for drivers and improving safety. The main feature of the current stage is “Human-machine Shared Control”. In order to meet the personalized driving needs of drivers, while ensuring safety, the driving habits and characteristics of drivers are fully considered, so that the decision-making and control results of the intelligent driving control system meet the expectations of drivers. Inverse reinforcement learning has shown good performance in personalized human learning and can learn the driving strategies of human drivers. However, many current methods of inverse reinforcement learning do not fully consider the interaction between vehicles. Therefore, this paper proposes a personalized ramp merging decision-making method based on maximum entropy inverse reinforcement learning, taking into account the interaction between vehicles. Based on driving style classification of human ramp merging data, targeted reward function forms are learned for different types of drivers to generate corresponding merging decision methods.