Vehicle Velocity Trajectory Optimization in Non-deterministic Environments

Optimally modulating a vehicle's speed profile through highway, urban, and suburban environments can result in pronounced fuel savings, particularly with heavy-duty vehicles. However, existing strategies for speed profile optimization traditionally rely on aspirational and often deterministic assumptions, which cease to be accurate in the presence of real-world features such as non-deterministic traffic and actuated signalized intersections. The research in this dissertation establishes a hierarchical Green-Light Approach Speed (h-GLAS) strategy for controlling vehicles traveling through non-deterministic highway, urban, and suburban environments. The h-GLAS strategy utilizes vehicle-to-infrastructure (V2I) communication to receive information about the route's topology, speed limits, and signal phase and timing (SPaT). For suburban environments that employ semi- and fully-actuated signalized intersections, past signal timing information is used forecast the future values of the signal phase lengths. This information is used to construct a desired velocity profile to be tracked by a semi-economic model predictive controller (MPC), which computes the optimal wheel force command for the vehicle. When traveling through highway environments, the desired velocity profile represents a globally optimal dynamic program solution, computed offline before the beginning of a trip. For urban and suburban environments, however, the velocity profile is constructed to allow the vehicle to arrive at upcoming intersections when the probability of a green signal is maximized. The semi-economic MPC minimizes a quadratic objective function, which reflects a trade-off between minimizing mechanical energy expenditure and braking effort, with tracking the supplied desired velocity profile. This MPC is unable to maintain vehicle following constraints without changing its convex nature, so a command governor (CG) is located downstream to maintain vehicle-following constraints efficiently. The CG modifies the MPC's control action by the minimal amount necessary to maintain safe vehicle following. Using the simulation packages from PTV VISSIM, the h-GLAS is validated against real signal timing algorithms within a stochastic traffic environment parameterized by real-world data. Simulation results show that the h-GLAS controller is capable of significantly reducing a vehicle's fuel consumption by 16%-26% when compared to a baseline control strategy traveling through the same suburban environment.