Deep Neural Network-Based Linear Quadratic Programming for Vehicle Path Tracking

Engineering notes

Key topics: self-driving vehicle, self-driving, 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

Path tracking is a crucial component for achieving autonomous steering in self-driving vehicles. This paper presents a novel path tracking algorithm that utilizes a deep neural network (DNN) to learn steering actions, enabling autonomous path tracking for vehicles. The algorithm trains the deep neural network using input and output data derived from traditional control strategies. These traditional controllers encompass full-state feedback control designed through linear quadratic programming (LQR) and feedforward control that takes road curvature into account. State variables from the path tracking system, road curvature, and lateral vehicle speed serve as training inputs for the deep neural network, allowing the trained model to replicate the steering behavior of the traditional controller. Simulation results demonstrate that the trained model successfully accomplishes the tracking task and exhibits steering behavior equivalent to the LQR control method with feedforward control.

Links and sources

• Official / arXiv page

Comments