SE PhD Final Defense: Anni Li

SE PhD Final Defense: Anni Li

TITLE: Optimal Maneuvering for Safe and Cooperative Autonomous Vehicles in Mixed Traffic

ADVISOR: Christos Cassandras SE, ECE, CISE

COMMITTEE: Chair: Emiliano Dall’Anese ECE, SE; Ioannis Paschalidis ECE, SE, BME; Sean Andersson ME, SE; Roberto Tron ME, SE

ABSTRACT: Optimal trajectory planning and control for Connected and Autonomous Vehicles (CAVs) in mixed traffic is a fundamental and challenging problem, especially when safety and optimization goals conflict with each other with the existence of human-driven vehicles (HDVs). While cooperative control of CAVs offers promising opportunities for enhancing traffic safety and efficiency, how to benefit from the presence of a limited fraction of CAVs in mixed traffic when CAVs must safely interact with uncontrollable HDVs remains an open question. The first part of this dissertation establishes a safe interaction between CAVs and HDVs in the lane-changing problem so that the best possible response of a CAV to actions by its neighboring HDVs is modeled and an optimal policy for the CAV to perform a safe maneuver is designed. This interaction is formulated using a game theoretic framework with an appropriate behavioral model for an HDV and an iterated best response (IBR) method is used to determine a Nash equilibrium. Moreover, CAV-HDV interaction can be eliminated or greatly reduced by CAV cooperation so that the optimal policy is independent of HDV behavior. The cost of CAVs in this policy is proved to be monotonic with respect to the length of the CAV-HDV gap when vehicle interaction starts. Second, considering the fact that the dynamics and human-in-the-loop control policies of HDVs are unknown and hard to predict in practice, this dissertation adopts an event-triggered Control Barrier Function (CBF) method for CAVs to ensure the safety between CAVs and HDVs, and implements it in highway lane-changing maneuvers. The employed event-triggered CBF method estimates HDV models online, constructs data-driven and state-feedback safety controllers, and transforms constrained optimal control problems for CAVs into a sequence of Quadratic Programs (QPs). It reduces the computation complexity, bridges the gap between optimal trajectory planning and real-time control, and provides flexibility for CAVs to execute lane changes while ensuring collision avoidance with HDVs. Beyond controlling CAVs, the final part of this dissertation develops a Cooperative Compliance Control (CCC) framework to incentivize HDVs to align their behavior with socially optimal objectives using a ``refundable toll" scheme so as to achieve a desired compliance probability for all non-compliant HDVs. A key challenge lies in the heterogeneous and unknown responses of human drivers to tolls, complicating controller design and compliance enforcement over the traffic network. To address this, the thesis employs Control Lyapunov Functions (CLFs) to adaptively correct crucial components of the compliance probability model online and demonstrates that the desired compliance probability for HDVs can be achieved. The CCC scheme can be applied at both the micro-level, to induce a desired acceleration/deceleration of non-cooperative vehicles, and the macro-level, to affect decision-making processes such as route guidance, thereby improving system-wide traffic efficiency.