Autonomous Merging onto the Highway Using LSTM Neural Network

Merging onto highways is a complex task for autonomous vehicles and can cause catastrophic crashes, especially when considering the high speeds of other vehicles in the highway. While methods like path tracking and reinforcement learning are commonly used to execute this task, they may not accurately imitate the behavior of an expert driver and simultaneously maintain accuracy, safety and passenger comfort. In this paper, we present an autonomous driver model that uses an LSTM neural network trained on data gathered from two expert drivers executing the maneuver in a driving simulator. We extracted five features from the dataset, and used them to train the neural network model. The model approximates the proper steering wheel angle at each time step and feeds it to a PD controller which sets the physical steering wheel angle by managing the input voltage of a DC motor mounted on the driving simulator’s steering rod. The neural network model achieved an R2 score of 0.82057 and successfully guided the vehicle to merge onto the highway and reach the second lane without collisions in multiple trials in the simulator. The results of this research shows the potential of using expert driver data and neural networks to improve the accuracy and safety of merging onto highways in autonomous vehicles. As autonomous vehicles become more prevalent in the automotive industry, it is essential to develop reliable and safe highway-merging strategies. Our model provides a promising solution to this challenge and can be further developed and integrated into autonomous vehicles to enhance their capabilities and ensure safe and efficient merging onto highways.