I would like to share my current work with you and get feedback from those who may have already developed a similar approach.
After abandoning my purely statistical analysis system, I decided to work on something more physics-based. Instead of looking only at past results, I built a small training program that models the ball and rotor like objects in space mechanics.
The idea is inspired by spacecraft trajectory calculations:
The ball behaves like a satellite orbiting.
The rotor behaves like a rotating planet.
By recording the timing of crossings, the software calculates angular velocities:
Ball: ωb=2π/Δt
Rotor: ωr=2π/Δtr
The ball slows down over time, so the app fits a linear or quadratic deceleration curve.
At the drop moment, it extrapolates the ball’s position and adds the measured τ_hit until the first diamond.
Finally, it combines the ball and rotor angles to predict the impact sector:θimpact=θball(t)−θrotor(t)
Extra features I added:
Input the actual winning number, the system recalibrates automatically.
A kernel file that learns from outcomes, like navigation corrections for spacecraft.
Automatic saving of spins and learning data.
A MISSED button to discard failed spins.
Support for a gamepad controller (ABXY + Start/Select), much easier than the keyboard when watching live spins or videos.
I would really appreciate if anyone here could share their thoughts:
Has anyone tried a similar trajectory-based approach?
Do you see any weak points or ideas for improvement?
Any suggestions for refining the learning kernel or adding extra parameters?
All acrobats, all programs from Jafco, Steve and a few others were made in the same algorithm… but all failed or at least did not give such an advantage which was expected
The amount of features that you need in such a program is never ending really if you plan to make it for all kinds of wheels, if it’s for only tilted ones it probably is faster,m but still there are lot’s of issues, that’s maybe why almost every single person i see here says “eye prediction is better than computers”, which is madness, but i see it when you end up with a simplistic app.
I already once stated somewhere else what is needed, so i will repeat it in a limited way.
Scatter peaks generator(chi-squared or similar) & sort, for various rotor speeds, diamond hits (if you have the ability to predict diamond hits with good accuracy), and other conditions
Ball deceleration generator from multiple ball samples
Ball ending speed approximation, the speed at the end that you get, is the most problematic, since many time’s it’s 0.5-0.8 of full rev(if the diamond that you clock at doesn’t get hit, and even when it does there are still some problems), since the ball slows a lot, it can be done either manually if you can leave the table for a moment and set it up, or automatically, which is the problematic one.
Drop point and diamond hit prediction is the hardest thing to do, on semi-tilted wheels, even quite complex methods fail, so if you want your predictor to work on such wheels, you will spend probably the most amount of time on it, leveled wheels are very rare to find so simplistic approaches that works for them specifically will be harder to apply, at that point you’re better off with tilted wheels.
Various controls, from rotor partial clocking to timing corrections for manual input.
Deceleration updates and monitoring, the deceleration changes over time, so you will have to find a way to monitor it(predict a specific number at a specific point for example, and monitor if the rotor is at a specified position at that time) as well update it.
Data collection & correlation, as well applicability in various conditions, the system should be applicable when you can’t see the whole wheel, as well any other inconvenience you might face in all kinds of conditions, it should be able to predict regardless(some older predictors would vibrate / zap and you had to observe the ball at a specific point where that occurred, that assumed that you could see what’s going on, and if you couldn’t, the systems would be useless for most of the spins).
Final features like diamond hit prediction, diamond isolation(prediction only when a specific diamond is predicted to be hit), timing isolation(predict at a specified speed for more stable results), prediction at any speed, accuracy and results tracking etc. might be included as well.
The conclusion is, to make a program that is applicable for a big portion of the wheels and not the most flawed ones the amount of problems that you can face is never ending, the methods used as well have to be tested and for each problem possible the best solution has to be found, which extends the process a lot, there’s not many sellers that sell anything good these days, and the fact as i already mention, that there are people on this forum, that think a human eye is better than a computer says, a lot that the computers that players make, for the most part aren’t the best…
I am few years into this, and to get it done is much more problematic than it seems like.
Main weak point is that hard to notice something what not was tested before…
And at all what I see it is only measuring speeds, nothing more… ? In my understanding that is easiest part of game - every can in one or other way to measure wheel and ball, but what next ?
Are big part in the spin in which ball behave very differently than what you expect…
Not thinks , but are facts - we can look to wining players and what they use…
I myself participated in making maybe hundred programs… and use in my actual play few of them…
But about what you talk, we talked here 20 years back…where are the results from these few years ?
Dude, you’re literally the only one repeating this stuff over, for you computers don’t work, that’s cool, cause most likely the one you’ve got SUCK, now let the other people talk if they wanna talk, and NO, Steven’s computers do work, he has the highest knowledge, and provided the most details about a lot of problems that you can face.
The results, once i am done i will record and post online, now the topic is about computer development, so if you think that computers don’t work, let the other people talk.
Thanks a lot for your answers. Since I couldn’t get video auto-tracking (YOLO, etc.) to work properly, I switched to manual tracking. I now consider my code more like a visual-ballistic assisted trajectory anticipator.
With manual input (clicking ball and rotor using my gamepad), the predictions are actually much more precise. The kernel is trained on past spins (real scores saved in my CSV folder), and so far it performs really well.
Next step: I’ll add a Chi-square test as an audit tool to measure average accuracy over 1,000 spins and share the results here.
One limitation: all online casinos I know call “no more bets” before the spin, which makes it tricky to practice. Does anyone know of an online casino where I could actually test my code during the spin?
Even with some chaotic outcomes after diamond hits, I still seem to hold an advantage.
Linda
There are few places that allow early bets after the dealer spins, but that’s not the point… We are mixing here computers and something like an advanced dealer signature, the methods are different, perhaps you don’t need any image detector nor physics for methods that you have to bet before they spin the ball.
The program i am making, these 2 methods are separate, and when it comes to new methods that predict before they spin the ball these require a lot of testing, and the edge is lower.
Update, at the end of each spin i had to input the number chosen by the roulette in the program. today i finally reached the 1000 spins. And i am happy to share here the result with you. The program hitted the right sector of 12 Numbers 87% right.
The next step for me is to miniaturise the program in a portable device to make it live cause after asking here I wasnt able to find an obline casino not stopping the bet before the spin.
Any one can Help to put the program in a device ? Thanks in advance
Hi, new member here. I just finished Mark Billing’s book, “Follow The Bouncing Ball”. A data point you might want to consider is “revs out”, or how many revolutions before the ball drops. Drop point is critical in predicitve models.
