[Home]PPP

Robo Home | Changes | Preferences | AllPages

Precise Prediction Pugilist

By PEZ with precise prediction code from rozu and Jim.

This is mainly a test on what would happen with Pugilist if I could fit precise prediction of it's reverse and forward positions in it.

Download: http://www.robocoderepository.com/BotDetail.jsp?id=2104

The first tests show that Pugilist 1.3.1 and other versions using prediction estimates and with flattening active really need precise(r) prediction to work. The GF1 spike was probably mainly from the inexactness of the prediction, especially approaching walls where it starts heavy WallSmoothing. But now I have a spike at GF-1 as well. I think this could be a bin smoothing problem, but my first attempts to compensate (assuming that smoothing gives the end bins a smaller share than they should) have failed. Stay tuned for some profile graphs here.

-- PEZ

Here:

Crazy profile if you ask me... -- PEZ

Yeah, one more implementation detail. Being able to predict my position accurately even when I'm approaching walls means I now dare go for the truth, purity and beauty in NoFearTheWalls. -- PEZ

Questions?

Looking at the profile... This isn't what you wanted. Is everithing working now? -- rozu

I think the predictor is working. But there must be other issues. I'll release it anyway. -- PEZ

And it's out. None of my tests show that it works though... -- PEZ

I think the line maxTurning = ... should be calculated before the line predictedVelocity = ... because maxTurning is a function of predictedVelocity a tick before (look here: FuturePosition). -- rozu


Robo Home | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited April 18, 2004 11:41 EST by PEZ (diff)
Search: