Mixed Platoon Control under Noise and Attacks: Robust Data-Driven Predictive Control and Human-in-the-Loop Validation

TL;DR

This work tackles robust control of mixed CAV/HDV platoons under noise and adversarial attacks by integrating data-driven reachability with DeeP-LCC in a tube-based predictive framework (RDeeP-LCC). It constructs a matrix zonotope over-approximation $\mathcal{M}_{ABHJ}$ from data, derives a stabilizing gain $K$, and employs Hankel matrices to enable online prediction while accounting for uncertainty via reachable sets. The system is decoupled into nominal and error dynamics, with the error reachable set propagated to tighten nominal constraints and a receding-horizon optimization computes the nominal input, which is augmented by a tube-based feedback. Human-in-the-loop experiments validate significant improvements in driving safety, tracking accuracy, energy efficiency, and comfort under both state-independent and state-dependent attacks, demonstrating practical robustness and real-time applicability.

Abstract

Controlling mixed platoons, which consist of both connected and automated vehicles (CAVs) and human-driven vehicles (HDVs), poses significant challenges due to the uncertain and unknown human driving behaviors. Data-driven control methods offer promising solutions by leveraging available trajectory data, but their performance can be compromised by noise and attacks. To address this issue, this paper proposes a Robust Data-EnablEd Predictive Leading Cruise Control (RDeeP-LCC) framework based on data-driven reachability analysis. The framework over-approximates system dynamics under noise and attack using a matrix zonotope set derived from data, and develops a stabilizing feedback control law. By decoupling the mixed platoon system into nominal and error components, we employ data-driven reachability sets to recursively compute error reachable sets that account for noise and attacks, and obtain tightened safety constraints of the nominal system. This leads to a robust data-driven predictive control framework, solved in a tube-based control manner. Human-in-the-loop experiments demonstrate that the RDeeP-LCC method significantly improves robustness against noise and attacks, while enhancing tracking accuracy, control efficiency, energy economy, driving comfort, and driving safety.

Figures (6)

• Figure 1: Schematic for the mixed platoon under the influence of noise and attacks. The noise and attacks affect the uplink and downlink of the cloud control platform, respectively.
• Figure 2: Schematic of the proposed RDeeP-LCC method for mixed platoons. In the offline learning phase (blue), the method utilizes collected data (yellow) to calculate the over-approximated system matrix set $\mathcal{M}_{\hbox{ABHJ}}$, derive a data-driven feedback control law $K$ to ensure stability for all possible systems, and generate the Hankel matrices. In the online control phase (green), the RDeeP-LCC solves for optimal control input for the CAV in a receding horizon strategy. Specifically, the system is decomposed into the error system and the nominal system. Using $\mathcal{M}_{\hbox{ABHJ}}$ and $K$, the method recursively derives the data-driven reachable set of error states, and then subtracts this set from the constraints of the original system to obtain a more compact nominal system constraint. Then, the nominal control input $u_\mathrm{z}(k)$ is calculated using the standard DeeP-LCC under the compact nominal constraint. Finally, the actual control input of the CAV is obtained by combining the nominal control input $u_\mathrm{z}(k)$ with the error feedback control input $u_\mathrm{e}(k)$ in a tube-based control manner.
• Figure 3: Human-in-the-loop experimental platform for mixed platoon control.
• Figure 4: The fitting results of equilibrium spacings and equilibrium velocities for the two HDVs.
• Figure 5: Velocity errors in mixed platoon human-in-the-loop experiments for five control methods under state-independent attacks. The red profiles represent the velocity error between the HV and the CAV, the green profiles show the velocity error between the CAV and the HDV 2, and the purple profiles depict the velocity error between the HDV 2 and the HDV 3.

Theorems & Definitions (11)

• Remark 1
• Remark 2
• Definition 1: Interval Set althoff2010reachability
• Definition 2: Zonotope Set kuhn1998rigorously
• Definition 3: Matrix Zonotope Set althoff2010reachability
• Remark 3
• Lemma 1
• Lemma 2
• Lemma 3
• Remark 4