Lane Switching

Overview

For this project, I built a planning stack for a highway driving agent that must perform safe, efficient lane changes to reach a target lane amid mixed traffic. The car is modeled with kinematic bicycle dynamics and perceives its surroundings through a simulated 2D lidar, so it has to plan and react without knowing the other drivers’ exact policies ahead of time — surrounding traffic follows the Intelligent Driver Model (IDM), a car-following model for longitudinal driving behavior.

Performance is judged first and foremost on collision avoidance and reaching the goal lane, then on secondary factors like risk (time-to-collision), maneuver speed, lane compliance, and ride comfort.