This Self-Driving Go-Kart Doesn’t Need Any Fancy Sensors to Get Around a Track

When you drive your car, your only real sensory input from outside the vehicle is coming through your eyeballs. And yet, you’re able to cruise down the road without accidentally hitting any mailboxes or pedestrians the vast majority of the time. So, why is it that self-driving cars need a whole suite of sensors and, even with those, can’t seem to navigate public roads without getting confused? The short answer is that computers don’t see or interpret images they way we do, so we give them more data. But Austin Blake proved that isn’t always necessary when he built this go-kart that can get around a track with just a camera.

To be fair to the other self-driving cars of the world, public roads are much, much more difficult than specially designed tracks with edge markings that are easy for computers to identify. The latter is what Blake used for this project after learning some important lessons with his previous self-driving mini Tesla project.

We covered that project a year ago and while it was impressive, the self-driving didn’t go quite as well as Blake hoped. The mini Tesla was supposed to be able to drive around a park walking path, but frequently struggled with the task.

So, Blake went back to the drawing board and this go-kart is the result. Unlike that park walking path, this track inside Blake’s workshop has very clear markers designating the edges of the road. During training, a machine learning model looked at the track while Blake navigated the course. As it watched, the model also monitored the steering wheel angle and was able to associate angles with the track it could see.

To let the go-kart try to reproduce Blake’s laps on its own, Blake gave it three Arduino Nano boards. The first just acts as a communication bridge between the laptop running the model and another Arduino that controls the steering. It passes along the steering angle calculated by the model. The second Arduino takes that steering angle and looks at the current steering angle (via a potentiometer) to determine how much to turn the wheel. It turns the wheel through a Cytron motor drive. The third and final Arduino controls the go-kart’s throttle with an analog signal going to speed controller similar to that of an RC car.

That proved to be a winning formula, as the go-kart was able to circle the tiny track just by looking ahead with a camera — no additional sensors necessary, other than the potentiometer that monitors the kart’s actual steering angle.