Passing Policies
Varying racing-line deviation penalties to study passing decisions under uncertainty, modeled as an MDP.
AA 228 introduced decision-making under uncertainty from a computational perspective — the tools for building autonomous and decision-support systems. Autonomous racing is a rich test bed: it pushes the limits of the vehicle while forcing decisions in highly dynamic situations. For a passing scenario, I explored how policies changed as I varied the penalty for deviating from the racing line, on a non-banked oval modeled as a Markov Decision Process with two vehicles (ego and other). A higher deviation penalty produced more lateral maneuvering and extra laps to pass; a lower penalty produced higher speeds but more passing in unsafe zones — a clean illustration of how objective design shapes behavior.