Stop! before you dismiss me as completely crazy, this is more of a proof of concept thing. Now, as i was browsing the youtubes, i watched an FRC team reveal, which demonstrated how their auto was compiled and created completely by machine learning, i was curious if this is even possible in robotc, or in pros, and if the cortex has enough computing power to not take several decades to get a successful autonomous. If anyone has any idea of how to start, or any examples in C, it’d be greatly appreciated. any help is appreciated and i believe that this would be the first of its type in robotc and vex. Cheers in advance!
Yes. Here’s a recent thread:
https://vexforum.com/t/artificial-intelligence/40340/1
Some of the basic concepts are covered in the discussion. Mostly, there’s little reason to do it, but if you choose to do it anyway, you can compute the learning with off-cortex computers, and run the resulting model on the cortex itself. Other similar ideas are discussed.
There are a ton of different types of machine learning. Did they say which methods were used and why? At work we have been doing a bit of random forest work lately on finding fraud.
This page lays out the various choices of algorithms used in machine learning. I could see some linear regression models for path planning coupled with decision trees.
http://machinelearningmastery.com/a-tour-of-machine-learning-algorithms/
Vision processing would definitely benefit from some ML to sort out what the sensors see.
ANy more info you had to share would be awesome
So, currently i’m using robotc virtual worlds to test autos built by another program put into robotc (2 points woo), its a bit of a hodge podge but it got it to about 20 cycles per minute (cycle is a 15 second auto)