Improving Lookahead Method

[CraftinCubing]

Hello everyone. I am working on a program that will act as a timer, with also a heavy emphasis on being a Trainer for various techniques.

I'm not anywhere close to being finished, but I needed some input for some things I would be testing and working on.

In your opinion, what method works best for improving your ability to look-ahead (during F2L)?
- With this question, it is assuming the solver already possesses the ability to solve any individual F2L case without looking at it.
- It is also assuming that techniques you learn such as reducing cube rotations and reducing number of turns are implied within the methods

If you would like to go one step further, I am eager to see some responses for suggestions on what you think the best way a program could help someone improve on their look-ahead.

 

[efattah]

Simple; connect to the webcam. Have the webcam take video of the solve as it happens. Use a machine vision library and you can isolate pauses in the solve. The program will see how smooth the solve is. More pauses = bad lookahead. In this fashion you can rate the lookahead. A person can then turn slower and they will get a higher lookahead score because their solve will be smoother, with fewer pauses, even though the solve time may be worse. Then, they can turn faster. You could even score the solve as [lookahead score]/[solve time].

You probably need 60fps to get accurate image recognition. Technically machine vision on a cube should be comparatively easy as the cube has fixed colors, much like a green screen in studio filming.

If image recognition is daunting, you might even be able to get by with just using image deltas. Convert the image to grayscale, then subtract the frame from another frame that occurred 0.1 seconds earlier. The sum of the subtraction (absolute value) gives the amount of change between the frames. Change will be minimal during solves. You can calibrate it with the user by running one solve with a pause in the middle. This would be the easiest method. A smooth solve has constant deltas between frames. If you don't want to interpret the deltas, you can graph the deltas and the inverse of the standard deviation would be the lookahead score. Less standard deviation in the frame-deltas means smoother solves.

I used this method to track moving objects in video for robots, years ago.

 

[WombatWarrior17]

I don't think that one method is better than all of the others, I think it's a mix of several methods that really help lookahead.

If you do all of the methods that you have listed in the poll (and possibly more), your lookahead would improve faster than if you were to just use one.