Here's a small self driving car project.
- It is end-to-end - straight from image to motor actions.
- Read the blogpost form writeup here.
- I did this in 2017 (!), and now I am kinda expert in autonomous navigation and I see that this has aged a bit (felt cute might delete later).
- It still has a bunch of value, and, please excuse me patting myself on the back, is still quite cool.
- What's also cool, is that I hacked a small RC car toy instead of buying a premade kit
- It runs a neural net that steers the car to stay on the road made from white pieces of paper.
- In 2017 I thought that's impressive that it is possible at all to pull it off as side project in the proverbial garage'n'basement circumstances.
- Now it's 2023 and people are driving real cars on the road in a simliar fashion, end to end using open source projects 🤯.
Anyway, here's how it works:
So the main steps are:
- Teleop the car and thus gather training data: pairs of
(image, steering inputs from human) - Train neural network on those pairs
- Have the trained neural network control the car
The repo doesn't have requirements.txt because I didn't know about it back at the time. xD I guess I will fix it at some point.
It's also untyped, mostly undocumented. But, the code is clean and short, and that's really cool I think.
It was python2.7.