jwdunn1 / carnd-pid-parameter-optimization Goto Github PK

Hello James - I stumbled across your repo when looking into Twiddle. I think it would be interesting and useful if you provided more context for this line in the README:

In the discussion forum, Andres Castano observed in 2012, that the Twiddle algorithm might be improved some orders of magnitude simply by adjusting the increase/decrease amount from 10% to 72.4%.

I guess this is the same Andres who worked at the robotics division of NASA / JPL. I see he's posted other writeups (like this one) based on Udacity CS373 forum discussions. So I presume he'd have no objection to you do something similar and quoting his posts or the whole thread here for reference.

If moving from 10% to 72.4% leads to such an improvement to Twiddle, it'd be interesting to see a bit more explanation of why this is so - and perhaps some insight into why Sebastian Thrun might have missed this previously. 72.4% sounds very exact, i.e. not just some ball-park figure, it'd be interesting to know how Andres or you came up with this value.