Скачать или смотреть Self-Driving Agent in the Video Game, "TORCS", with Behavioral Cloning

Here, I have implemented a self-driving agent in the video game called TORCS with Behavioral Cloning. In essence, the behavioral cloning agent, as the name would suggest, is supposed to imitate the driving behavior of a human expert by making sense of his own driving experiences collected from the environment. Extracting the necessary information, such as the distance between the car and the road marks, from an image with conventional image processing methods is a tedious task and a time consuming process. Thus, in this project, the states of the agent are represented as a stack of gray scale frames which are composed of (66 x 105) pixels. The frames are fed to the convolutional neural network and the network gives an action at a time. In addition, as the state representation, last 4 consecutive states were also utilized in order to provide speed information to the agent.