Improving safety in mixed traffic: A learning-based model predictive control for autonomous and human-driven vehicle platooning

TL;DR

This work tackles safety in mixed-traffic platooning by modeling human-driven vehicle behavior with a hybrid ARX+GP approach and embedding HV uncertainty into a model predictive control framework. The HV model combines a first-principles delay-inclusive transfer function with a Gaussian process correction, enabling a GP-MPC that uses the HV mean and variance to enforce a chance-constrained safe distance. Simulation results show the GP-MPC outperforms a baseline MPC, delivering larger safety margins and higher throughput, particularly in emergency braking scenarios (e.g., increasing HV–AV minimum distance from about $20.7$ m to $23.1$ m). The proposed framework advances practical, uncertainty-aware control for mixed-traffic platooning, though it relies on offline GP training and could benefit from richer driving data and scalable GP methods for real-world deployment.

Abstract

As autonomous vehicles (AVs) become more common on public roads, their interaction with human-driven vehicles (HVs) in mixed traffic is inevitable. This requires new control strategies for AVs to handle the unpredictable nature of HVs. This study focused on safe control in mixed-vehicle platoons consisting of both AVs and HVs, particularly during longitudinal car-following scenarios. We introduce a novel model that combines a conventional first-principles model with a Gaussian process (GP) machine learning-based model to better predict HV behavior. Our results showed a significant improvement in predicting HV speed, with a 35.64% reduction in the root mean square error compared with the use of the first-principles model alone. We developed a new control strategy called GP-MPC, which uses the proposed HV model for safer distance management between vehicles in the mixed platoon. The GP-MPC strategy effectively utilizes the capacity of the GP model to assess uncertainties, thereby significantly enhancing safety in challenging traffic scenarios, such as emergency braking scenarios. In simulations, the GP-MPC strategy outperformed the baseline MPC method, offering better safety and more efficient vehicle movement in mixed traffic.

Figures (7)

• Figure 1: Platoon of $N_a$ autonomous vehicles (AVs) depicted as $\{A^1, A^2, \cdots, A^{N_a}\}$ and a following human-driven vehicle (HV) $H$. The AVs in the platoon can exchange information through a bidirectional communication topology. This cooperative behavior contrasts with a single AV--HV scenario in which the traffic ahead is not cooperative. In an AV platoon, AVs can synchronize their behaviors to create safer interaction between the last AV and the trailing HV.
• Figure 2: One of three distracted drivers in a controlled experiment, designed to gather data for estimating the HV model in a Unity driving simulator.
• Figure 3: Performance of the ARX+GP model evaluated on one testing dataset. The HV velocity predictions of the ARX and ARX+GP models were compared by plotting them alongside the measured velocities and twice the standard deviation (2$\sigma$) determined using the GP model. The results highlight that the ARX+GP model significantly enhanced the fit of velocity curves compared with the standalone ARX model. On average, the ARX+GP model improved modeling accuracy by 35.64% in terms of the RMSE compared with the ARX model.
• Figure 4: Results of a constant velocity-tracking simulation utilizing GP-MPC. Sequentially from top to bottom, the plots illustrate the velocity response, position trajectories, and inter-vehicle distance. These graphs demonstrate the stable performance of GP-MPC in achieving and maintaining the desired platoon behavior under constant velocity conditions.
• Figure 5: Results of an emergency braking simulation using the baseline MPC. The figures represent, in descending order, the velocity response, position trajectories, and inter-vehicle distance.