Autonomous Self-Driving Agent

Overview

Architected and implemented an autonomous vehicle agent capable of navigating complex procedural road environments in real-time. The system processes direct visual streams to extract road contours, track lane curvature, and execute steering and throttle commands via a closed-loop controller.

Perception Pipeline & Algorithms

• Edge & Contour Filtering: Custom OpenCV pipeline isolating road surface boundaries and terrain contrast under variable lighting conditions.
• Spatial Clustering (DBSCAN): Filters out noise, outliers, and environmental debris from visual feature clouds.
• Robust Line Fitting (RANSAC): Fits accurate polynomial curves to lane boundaries even in high-noise visual environments, achieving 99.97% perception reliability.
• Trajectory Control: Multi-stage PID controller with genetic algorithm gain tuning, maintaining center-lane stability across extreme curvature.

Source Code & Verification