Hacker News Reader

Comment by ipunchghosts | Hacker News Reader

X is in the range 0, 255. They don't show code converting it to a much saner range for the network they've chosen. Is the full source somewhere?

View on HN

Have you already looked at the pickled data? Because it looks like the model is outputting a single label value out of 256 labels; depending on the training data (steering angle) and how it is represented in the data (signed float or integer?), each one of those 256 learned should (?) be similar - I think.

Again, I'm not an expert. Or - maybe it is outputting a number 0-255, and then taking that number and converting it (and maybe other operations) into values suitable for the servo on the car (perhaps centered around 0 - so -128 to 127 or something like that - then scaled for servo PPM width or whatever values needed)...

All guesses, of course.

The input values are image arrays 120x160 pixels with 3 channels for red,green,blue. The values range from 0-255 and are not normalized before they are fed into the convolution layer. I found this did not make a difference.

The output of the model is a single real number between -90(left) and 90(right). I believe a better approach would be to bin the outputs and use a classifier. This way you'd know when the model was getting confused (ie, approaching a perpendicular line.

Also the full repo is mentioned (I think the article is just a general highlighting of the full repo):

https://github.com/wroscoe/donkey

Reply on news.ycombinator.com