An Integrated Framework for Autonomous Driving Planning and Tracking Based on NNMPC Considering Road Surface Variations
Lists
Only you see this
Abstract
The planning and control subsystem is essential to the autonomous driving system. The prevailing algorithmic architecture follows a hierarchical design: the planning subsystem generates a safe trajectory based on the surrounding environment, and the control subsystem guides the vehicle driving along this trajectory. However, due to the limitations of the vehicle dynamics, in specific emergencies or on low-adhesion road surfaces, the generated trajectory may not be compatible with the vehicle's tracking ability, potentially resulting in difficulties in trajectory tracking or overly conservative trajectory generation. This paper proposes an integrated planning and control framework based on the artificial potential field and neural network vehicle model to address the above issue. Utilizing a neural network to model the vehicle not only ensures model accuracy but also empowers it with the capability to adapt to diverse road surfaces. The traffic environment is represented as an artificial potential field model using improved potential field functions. In a unified optimization framework, the above two aspects are combined to generate safe and efficient vehicle control commands directly. The test results under urgent scenarios and low-adhesion road surfaces indicate that the proposed method consistently keeps the vehicle within safe and stable boundaries while achieving driving objectives. It produces smoother driving trajectories and smaller lateral accelerations compared to hierarchical algorithms.
BibTeX
@article{gao2025integrated,
author = {Zhenhai Gao and Wenhao Wen and Yang Xing and Antonios Tsourdos},
title = {{An Integrated Framework for Autonomous Driving Planning and Tracking Based on NNMPC Considering Road Surface Variations}},
journal = {IEEE Trans. Intell. Veh.},
year = {2025},
doi = {10.1109/tiv.2024.3418951},
}Loading related content...