Vehicle Longitudinal Control under Autonomy, Connectivity, and Mixed-flow Traffic

Thumbnail Image
Zhou, Anye
Peeta, Srinivas
Associated Organization(s)
Supplementary to
Rapid advances in connectivity and vehicle automation technologies are enabling the development of connected and autonomous vehicles (CAVs), which provide promising opportunities to improve the current transportation system. The longitudinal control of CAVs can leverage the information obtained using connectivity and sensor measurements to optimize vehicle trajectories to improve travel comfort, safety, fuel economy, and traffic efficiency. However, critical challenges related to vehicle longitudinal control from the current stage to the long-term future can reduce CAV benefits to the transportation system: (i) propagation of traffic congestion due to inappropriate autonomy design; (ii) connectivity disruptions and traffic oscillations induced by uncertain human driving behaviors in mixed-flow traffic of CAVs and human-driven vehicles (HDVs); (iii) uncertainties in vehicle dynamics (e.g., rolling resistance, air drag, gravity effects induced by road grade, and powertrain lag) which prevent vehicles from executing planned trajectories; and (iv) communication failures triggered by unreliable communication technologies and falsified information injection caused by cyberattacks in the communication process which significantly compromise safety and mobility. The dissertation seeks to develop novel, effective, and efficient solutions to broadly deepen our understanding of vehicle longitudinal control for the transition from the current-stage of autonomy to the pure CAV environment in the long-term future, and systematically address the associated challenges in real-world operations. First, starting with the current stage when commercially-available adaptive cruise control (ACC) systems (in Level 2 autonomous driving) exacerbate traffic congestion (i.e., string unstable), this dissertation develops a cost-effective control mechanism that enables congestion mitigation without altering the existing control algorithms inside ACC systems. The proposed mechanism includes a trajectory shaper (TS) and a linear acceleration/deceleration relaxation (LA/D-R) technique. The TS modifies the trajectory information of the predecessor vehicle, which induces string stable car-following (CF) behavior from a string unstable ACC system. The LA/D-R alters the original acceleration/deceleration limit of an ACC system using a spacing error-dependent limit, which ensures smooth vehicle acceleration/deceleration profiles and mitigates the negative impacts of the original acceleration/deceleration limit. The proposed control mechanism can reduce the costs of tuning and revising existing ACC algorithms and/or designing new ACC algorithms. Second, by acknowledging the challenges that HDVs would face in mixed-flow traffic in the near-term future, the dissertation first presents a driving simulator study that provides insights into the heterogeneity in HDV CF behavior when following CAVs with different control settings under different traffic congestion scenarios. Mediation analysis indicates that human-like CAVs inducing the least difference in CF behaviors with HDVs can trigger the smoothest CF response from HDVs. Parameter estimation using particle filters indicates variable HDV responses towards different stimuli from CAVs and traffic congestion. The behavior evolution evaluated using extended Kalman filter unveils a gradual decaying capability in dampening speed fluctuations from HDVs. These findings can serve as a fundamental basis for developing intelligent longitudinal control to improve control performance in mixed-flow traffic. Then, the dissertation proposes a cooperative control strategy to mitigate the effects of connectivity disruptions and traffic oscillations induced by HDVs. This strategy consists of a deep neural network (DNN)-based HDV number estimator and kinematic state predictor, and multi-anticipative car-following control that combines a model-based extended intelligent driver model (EIDM) and a model-free deep deterministic policy gradient (DDPG). The DNN-based HDV number estimator and kinematic state predictor can accurately identify the number of HDVs between two CAVs and their kinematic states so that the connectivity disruption is compensated. The proposed multi-anticipative car-following control enhances traffic efficiency and string stability (outperforming the existing EIDM), and substantially improves sampling efficiency and convergence during training compared to DDPG. This study highlights the significant potential of hybrid approaches that integrate model-based and model-free methods to address various real-world challenges associated with vehicle longitudinal control. Third, based on the expectation of a pure CAV environment in the long-term future, the dissertation develops an adaptive smooth switching control strategy and a robust control strategy to sustain CAV operations under the impacts of communication failures and falsified information injection, respectively. The smooth switching control strategy integrates information flow topology optimization (to reduce the possibility of communication failures), controller parameter optimization (to ensure smoothness during control switching), and a Kalman predictor (to suppress measurement noise and predict kinematic states to reduce control switching and enhance smoothness). It enables smooth and string-stable CAV platoon operations under unreliable connectivity. The robust control strategy first applies a linear state and disturbance observer to compute estimated vehicle states to alleviate the impacts of falsified information. Then, a robust control law is used to suppress the disturbances induced by falsified information and inaccurately estimated vehicle states. A control barrier function-based control decision regulator is then devised to guarantee feasible acceleration/deceleration and safe spacing during operations. These three components enable safe, smooth, and computationally efficient CAV operations under connectivity with falsified information. Last, under the assumption of mature connectivity and vehicle automation technologies, virtual-platooning control and traffic flow regulation are used to devise a cooperative signal-free intersection control strategy for CAVs. First, virtual-platoon coordination is performed to optimize the passing sequence and platoon formation so that intersection throughput is maximized and structural controllability is achieved to enable vehicles to maintain safe spacings. Then, the virtual-platooning control implements a distributed adaptive sliding mode controller for CAVs to cooperatively mitigate the negative impacts of uncertain vehicle dynamics to generate safe and trackable vehicle trajectories. Further, traffic flow regulation is performed to mitigate traffic spillbacks under high traffic demand using constrained finite-time optimal control. The proposed cooperative signal-free intersection control strategy coordinates vehicles approaching from different directions simultaneously, significantly improving traffic efficiency and mobility compared to traditional signalized intersection control. In summary, the dissertation deepens the understanding of challenges in CAV operations from the current stage to the long-term future, which can assist CAV manufacturers to design effective vehicle control and transportation managers in designing traffic management strategies to fully exploit the potential of CAVs to improve the transportation system. The proposed control strategies can significantly enhance CAV reliability when facing challenges in real-world operations, enabling them to be more reliable and appealing to consumers and helping the deliberate scale up of CAV adoption as the future unfolds.
Date Issued
Resource Type
Resource Subtype
Rights Statement
Rights URI