800 riderless bicycles

Page may contain affiliate links. Please see terms for details.

winjim

Smash the cistern
Somebody has been doing some research into teaching a neural net to ride a bike.

PDF: http://paradise.caltech.edu/~cook/papers/TwoNeurons.pdf

I haven't had a chance to read it yet but it does contain this rather lovely image of the path taken by a riderless bike, pushed until it falls over, repeated 800 times.

DURHC9VXUAAwu_U.jpg
 
Randomness in practice.
 

Globalti

Legendary Member
But it's not random is it? If it was, the picture would be just a mass of curly black lines like a hipster's beard, but there are clear patterns in the way the bike falls over.
 

C R

Guru
Location
Worcester
But it's not random is it? If it was, the picture would be just a mass of curly black lines like a hipster's beard, but there are clear patterns in the way the bike falls over.

It is random in the sense that for a given ride you can't tell how far it will go and which route it will follow. For the given experiment you would not expect the route map to cover the space uniformly, because you always launch the bike in the same direction.

What is quite interesting is the periodicity of the excursions, I wonder if the spacing of those relate to the wheel size. I guess I should read the paper.

Edit to add. Having quickly skimmed the paper, it is a simulation, so some ideal behaviour will be seen. I think they may have used Scilab for the simulation by the look of some of the graphs. I seem to remember some time ago seeing a Scilab demo that showed an unsteered bike wobbling.
 
Last edited:
U

User169

Guest
Strangely written paper. The figure isn’t referred to in the text of the paper. You only have the legend to go on.
 
Top Bottom