Car-following algorithm based on TD3 and CACC

For autonomous cars to drive safely and efficiently, they need to have a decent car following functionality. Many algorithms have been proposed to improve the car following functionality, but most of them focus at keeping similar speed instead of maintaining relative distances between vehicles, which makes these algorithms hard to deal with emergencies. Therefore, the paper proposed a hybrid algorithm based on twin delayed deep deterministic policy gradient (TD3) and cooperative adaptive cruise control (CACC) for car following task in autonomous driving. The algorithm is tested in an environment which simulates typical scenario in real world highway driving, and the performance is also measured when vehicle to vehicle (V2V) latency is also considered. The experiment shows that the algorithm keeps a decent dynamic relative distance between vehicles, which is crucial for collision avoidance in case of emergencies. It has a good performance and a great potential for real world applications.