Real-Time Control Using Convolution Neural Network for Self-Driving Cars

Engineering notes

Key topics: self-driving car, self-driving, end-to-end, planning, control. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

In this paper, we perform an Autonomous deep learning robot using an end-to-end system. The system operates as the controller for navigating and driving automatically. The deep learning robot used Convolution Neural Network (CNN). The CNN architecture is Mobile net with Softmax activation function. The Softmax activation function predicts the probability of steering angles. In the training phase, the CNN model learns from images and steering angles that are collected during the driving. In the testing phase, we apply the diversified environment to the trained CNN model. The CNN model accuracy is up to 85.03%. The results showed that the CNN is able to learn the diversified tasks of lanes and roads following with and without lane marking, direction planning and automatically control. Also, the CNN can replace the conventional PID controller.

Links and sources

• Official / arXiv page

Comments