SE PhD Final Defense: Andres Chavez Armijos

Title: "FROM SELFISH TO SOCIAL OPTIMAL PLANNING FOR COOPERATIVE AUTONOMOUS VEHICLES IN TRANSPORTATION SYSTEMS"

Advisory Committee: Christos Cassandras ECE, SE (Advisor); Ioannis Paschalidis ECE, SE, BME, CDS; Sean Andersson ME, SE; Roberto Tron ME, SE; David Castañón ECE, SE (Chair)

Abstract: Connected and Automated Vehicles (CAVs) have the potential to revolutionize transportation efficiency and safety through collaborative behavior. This dissertation explores the challenges and opportunities associated with achieving socially optimal cooperative maneuvers, using the case study of cooperative lane-changing to showcase the significance of cooperativeness. Cooperative lane-changing serves as an ideal testbed for examining decentralized optimal control, interactions with uncooperative vehicles, accommodating diverse human driving preferences, and integrating planning and execution processes.

Initially, the research focuses on scenarios where all vehicles are cooperative CAVs, leveraging their communication and coordination capabilities. Decentralized optimal control problems are formulated to minimize energy consumption, travel time, and traffic disruption during sequential cooperative lane changes, balancing individual vehicle objectives with system-level goals.

The dissertation then extends the analysis to mixed-traffic scenarios involving uncooperative human-driven vehicles (HDVs). A novel approach is developed to ensure safety assurance, combining optimal control with Control Barrier Functions (CBFs) and fixed-time convergence (FxT-OCBF). Robust methods for handling disturbances from uncooperative vehicles are introduced, enhancing the resilience and dependability of cooperative lane-changing maneuvers.

To address the complexities of CAVs interacting with HDVs exhibiting diverse driving preferences, an innovative online learning framework is presented. Safety preferences are characterized using parameterized CBFs, and an extended Kalman filter dynamically adjusts control parameters based on observed interactions, enabling real-time adaptation to evolving human behaviors.

The proposed methodologies bridge the gap between high-level planning and low-level control execution, facilitating safe and near-optimal cooperative maneuvers. Comprehensive analysis demonstrates improved traffic throughput, reduced energy consumption, and enhanced safety compared to non-cooperative or reactive approaches. This research lays the foundation for deploying CAV technologies that prioritize social optimality while addressing uncertainties in mixed-traffic settings, ultimately paving the way for safer and more efficient transportation systems.