CNN-based Adaptive Intelligent Driving and Algorithm Optimization

Autopilot is the latest application of artificial intelligence technology in the automotive field. This is the future trend of automotive development and is seen as a beacon to lead the new automotive industry revolution. There are many research directions in autopilot technology, including Direct Perception, indirect Mediated Perception, and End-to-End Control. this article mainly uses End-to-End Control as the core research method. We use the general neural network and ALexNet, Nvidia-net in CNN as the control side to do training work and evaluate the performance of these three neural networks.We then proposed an improved solution for the Nividia-net algorithm.The flask framework is used to build the server as the control side to analyze the characters of network models and allow the controlled side to send the control information. In order to reduce costs, this research uses a virtual racing game as the controlled side. At the same time, this research collects the information of frontal pictures, steering wheel angles, brakes, throttle, and speed of the racing car through the recording function of the virtual racing game, which is used as training data. Through the above methods, we completed the case of training, running and testing.