A Parameter Privacy-Preserving Strategy for Mixed-Autonomy Platoon Control

TL;DR

This paper integrates a parameter privacy filter into LCC to protect sensitive car-following parameters and demonstrates the benefit of such an approach in networked systems, i.e., by applying the privacy filter to a preceding vehicle, one can also achieve a certain level of privacy for the following vehicle.

Abstract

It has been demonstrated that leading cruise control (LCC) can improve the operation of mixed-autonomy platoons by allowing connected and automated vehicles (CAVs) to make longitudinal control decisions based on the information provided by surrounding vehicles. However, LCC generally requires surrounding human-driven vehicles (HDVs) to share their real-time states, which can be used by adversaries to infer drivers' car-following behavior, potentially leading to financial losses or safety concerns. This paper aims to address such privacy concerns and protect the behavioral characteristics of HDVs by devising a parameter privacy-preserving approach for mixed-autonomy platoon control. First, we integrate a parameter privacy filter into LCC to protect sensitive car-following parameters. The privacy filter allows each vehicle to generate seemingly realistic pseudo states by distorting the true parameters to pseudo parameters, which can protect drivers' privacy in behavioral parameters without significantly influencing the control performance. Second, to enhance the reliability and practicality of the privacy filter within LCC, we first introduce an individual-level parameter privacy preservation constraint to the privacy filter, focusing on the privacy level of each individual parameter pair. Subsequently, we extend the current approach to accommodate continuous parameter spaces through a neural network estimator. Third, analysis of head-to-tail string stability reveals the potential impact of privacy filters in degrading mixed traffic flow performance. Simulation shows that this approach can effectively trade off privacy and control performance in LCC. We further demonstrate the benefit of such an approach in networked systems, i.e., by applying the privacy filter to a preceding vehicle, one can also achieve a certain level of privacy for the following vehicle.

Figures (12)

• Figure 1: Overview of the proposed parameter privacy filter for the mixed-autonomy platoon, where blue vehicles represent CAVs, gray vehicles represent HDVs, and the black vehicle denotes the head vehicle. Some HDVs are equipped with a privacy filter $\Gamma(\cdot)$ to protect the privacy of sensitive car-following parameters. The proposed filter enables HDVs to distort their reported states to pseudo states, thereby protecting sensitive parameters. The trade-offs between privacy and control performance are analyzed.
• Figure 2: Structure of the neural network-based randomizer.
• Figure 3: Implementation scenarios of the parameter privacy filter in platoon control: (a) neither vehicle $i$ nor its preceding vehicle $i-1$ has implemented the privacy filter, (b) only vehicle $i$ has implemented the privacy filter, (c) only vehicle $i-1$ has implemented the privacy filter, and (d) both vehicles have implemented the privacy filter.
• Figure 4: Simulation results for the scenario in Fig. \ref{['fig: privacy scenarios']} (b) using parameter privacy filter with a discretization-based randomizer. (a) and (b) show the normalized RMSE of the parameter estimator over time for the driving scenarios of sine-like velocity disturbances and emergency braking, respectively, with $I_0=0.2$ and a Gaussian additive noise intensity of $0.2$. Note that the higher the RMSE, the better the privacy protection. (c) illustrates the normalized RMSE error of the parameter estimator in scenarios with various levels of information budget $I_0$ for two driving scenarios. (d) represents average total distortion in scenarios with various levels of information budget $I_0$ for two driving scenarios.
• Figure 5: Simulation results for the scenario in Fig. \ref{['fig: privacy scenarios']} (c) using the parameter privacy filter with a discretization-based randomizer. (a) and (b) indicate the normalized RMSE of the parameter estimator as a function of time for the driving scenarios of sine-like velocity disturbances and emergency braking, respectively, with $I_0=0.2$ and a Gaussian additive noise intensity of $0.2$. (c) shows the normalized RMSE error of the parameter estimator versus the level of information leakage $I_0$ for two driving scenarios.

Theorems & Definitions (3)

• Definition 1: Car-following Model
• Remark 1: Safety considerations
• Definition 2: Head-to-Tail String Stability jin2014dynamics