Review of Lane Change Systems Based on Reinforcement Learning

The decision of lane change problem is one of the sophisticated and accuracy demanding task in ADAS development process. Lane change task can be characterized as non- linearity with high fraction of uncertainties. This is caused by dynamic changeable traffic environment and thus one of state of the art approach to solve this problem is the use of machine learning technic. In this work lane change systems based on reinforcement learning methods were observed. These lane change systems are provided by specific modification of deep Q-network (DQN), deep deterministic policy gradient (DDPG). In the process of consideration descriptions and the comparison of proposed methods are given from position of Average Velocity and Collision Times.