Converting a human-driven electric racecar into an autonomous vehicle was a system-integration challenge. Perception, planning and control all depended on reliable hardware and clear interfaces.
Vehicle conversion
The car was equipped with a VLP-16 lidar, a wide-view monocular camera and a long-range stereo camera. The conversion also added onboard compute, active steering actuation and emergency braking.
Perception
I trained a YOLO-based camera model to detect the cones that defined the track. Its output became part of a larger autonomy stack, where perception had to remain useful under real track conditions rather than only on a curated dataset.
What shaped my work
This project reinforced a systems-first approach to robotics: make interfaces observable, test assumptions at subsystem boundaries and treat safety, hardware and software as one product.