Originally published at: Try this machine learning simulation to train a self-parking car | Boing Boing
…
About 20 years ago I was told of the PhD thesis of a student of Jerry Marsden’s, the mathematics of parking a five-stage tractor-trailer. It included the math, a computer simulation, and a toy truck that would indeed park on command. I’m not finding it now although this is fairly close.
This being a genetic algorithm, I keep waiting for it to decide that the winning move is to ram another car at top speed such that it flips upside down and lands neatly on the parking spot. Or some other wacky non-solution that somehow registers as a solution.
I’ve always put my (in)ability to successfully reverse with a trailer down to some sort of Heisenburgian thing. I can never know where the trailer is and where it’s meant to go at the same time. It’s probably quantum.
I put my inability to consistently reverse a car down to not learning young enough. I know some farm-born people who had no trouble passing their heavy vehicle license test as soon as they reached the minimum age, because they had to drive tractors around the farm from whenever they could reach the pedals.
I’m noticing that the closest thing to a solution the GA is managing is to do a reverse U-turn through the parking spot, so they may be on their way to that…
I wish it saved the best car… like, there was a generation a ways back that had a loss of 0.2, never to be replicated again.
The current best car, after accidentally leaving this on for 18 hours:
This topic was automatically closed after 5 days. New replies are no longer allowed.