EVC go-kart vision

Problem

Purdue’s Electric Vehicle Club builds an autonomous electric go-kart to race in the Autonomous Karting Series. I was on the three-person computer vision team, which turned the kart’s camera feed into track data for the team that planned the racing line.

What I did

• Wrote a Python and OpenCV pipeline that finds the edges of the track in each frame, using color filtering, edge detection and a top-down warp of the image, then fitting lines to trace the curves.
• Estimated the hidden edge on sharp turns, when only one side of the track is in view, from the visible edge and the track’s known width.
• Output the angle to each side of the track, which the racing team used to choose the line.
• Optimized it to run in real time on the kart’s onboard Jetson.
• Worked in a codebase shared across the club’s teams, using Git and Docker.

Result

• The team placed 1st of 9 teams from the US and Canada at the 2026 Autonomous Karting Series US National Grand Prix.
• The kart was also 1st in qualifying, with the fastest lap (1:03).

What I’d change

I’d get more involved with the other teams. The better I understood the rest of the kart, the better I could have tuned the vision system for it.