Data-Driven Cooperative Adaptive Cruise Control for Unknown Nonlinear Vehicle Platoons

Jianglin Lan

Abstract

This paper presents a novel data-driven control strategy for vehicle platoons that manages unknown nonlinear dynamics without explicit mathematical modeling. By leveraging a Semidefinite Program (SDP) formulation and $H_\infty$ performance criteria, the proposed method ensures string stability and precise tracking for both fully automated and mixed-autonomy convoys. Simulation results using the SFTP-US06 drive cycle demonstrate significant performance gains over traditional adaptive cruise control systems.

Introduction

Cooperative Adaptive Cruise Control (CACC) is a cornerstone of modern intelligent transportation, enabling vehicles to travel in tight "platoons" via vehicle-to-vehicle (V2V) communication. This proximity improves road capacity and fuel efficiency. However, real-world implementation is hindered by complex nonlinearities—such as aerodynamic drag, mechanical friction, and varying vehicle mass—that are difficult to model accurately. Traditional CACC often relies on simplified linear "point-mass" models or feedback linearization, which requires precise knowledge of physical parameters. This research addresses these challenges by developing a robust, model-free CACC system capable of handling unknown nonlinearities and unpredictable human-driven vehicles (HVs).

Methodology

The proposed approach bypasses explicit system identification by utilizing a data-driven framework. The unknown nonlinear dynamics of the platoon are represented as a discrete-time polynomial system. The controller "learns" from online data samples, including vehicle accelerations, spacing errors, and relative velocities.

The control design is formulated as a Semidefinite Program (SDP), incorporating $H_\infty$ performance criteria to ensure robustness against external disturbances like wind gusts or sudden lead-vehicle maneuvers. To maintain computational efficiency for larger convoys, the authors implemented a "virtual splitting" technique, where the platoon is divided into smaller sub-groups for parallel SDP solving.

Evaluation was conducted using high-fidelity simulations against the SFTP-US06 Drive Cycle, an aggressive high-speed benchmark. Vehicles were modeled with 1500kg mass and $\pm 10\%$ random parameter deviations to simulate real-world heterogeneity.

Key Results and Discussion

The data-driven CACC demonstrated significant improvements in both stability and tracking precision compared to classic ACC.

String Stability and Spacing

In a four-vehicle automated platoon, the proposed method maintained inter-vehicular distances remarkably close to the 20-meter target. In contrast, classic ACC exhibited "string instability," where oscillations in spacing grew as they propagated toward the rear of the platoon.

Inter-vehicular distance comparison between classic ACC (a) and the proposed data-driven method (b).

Velocity Tracking

Under the erratic speed profile of the SFTP-US06 cycle, the proposed method kept velocity deviations within a tight bound (approx. $\pm 2$ m/s). Classic ACC struggled significantly, with deviations reaching approximately $\pm 10$ m/s in some instances.

Velocity deviation comparison showing the tighter tracking bounds achieved by the proposed method (b) compared to classic ACC (a).

Mixed Autonomy Performance

The framework proved equally effective in mixed platoons. By treating human-driven vehicle behavior as an unknown nonlinear component, the automated vehicles at the front and rear successfully stabilized the convoy without needing to model human car-following logic.

Conclusion and Future Work

This research provides a unified, scalable framework for platoon control that thrives on data rather than complex physical modeling. While the current results are promising, future work will focus on incorporating hard safety constraints to prevent collisions during system failures and addressing real-world communication issues like latency and packet loss. Additionally, the requirement for an initial "safe" data collection period suggests a hybrid approach might be necessary for initial deployment.

Summary updated

BibTeX
@inproceedings{lan2024data,
  author = {Jianglin Lan},
  title = {{Data-Driven Cooperative Adaptive Cruise Control for Unknown Nonlinear Vehicle Platoons}},
  booktitle = {IET},
  year = {2024},
  doi = {10.1049/itr2.12556},
}
Figure