Precise Longitudinal Control of Automated Vehicles without Complex
Modeling Based on Physical Data 12-06-03-0020
This also appears in
SAE International Journal of Connected and Automated Vehicles-V132-12EJ
Precise controls of vehicle states are crucial to automated vehicles (AVs).
Traditional model-based AV control algorithms require complex modeling and
controller design, and their accuracy is still affected a lot by various
uncertainties. Latest data-driven controls such as artificial neural network
(ANN)-based controls can reduce modeling efforts but are usually subject to
robustness issues in unseen scenarios. This article proposes to combine a
data-driven control and a typical analytical model-formed control to achieve a
better AV longitudinal control performance with fewer modeling efforts. The
data-driven control can handle the complex modeling, calibration, and controller
design, and the analytical model-formed control can guide the direction of the
control with better predictability and robustness in unseen scenarios. The
proposed controller is experimentally implemented and validated using a real AV.
The performance is compared to the standalone data-driven controller and
analytic model-formed controller, and the experimental results demonstrate the
effectiveness and advantages of the proposed approach.