Bug within the initialization of the calibration

1

Hi. I think the initialization of the first guess is odd. If you take a look into the calibration simulation there are two lines:
guessThisTime.br.x = guessThisTime.bl.x + (Math.random() * 400) - 200;
guessThisTime.br.y = guessThisTime.bl.y + (Math.random() * 400) - 200;
which means that the bottom right becomes the initial values of the bottom left. This can lead to non convergence but at least the convergence becomes very slow. Attach
Maslow-serial(28).log (119.5 KB)
I think this is also done by the common calibration. If you take a look onto the initial 3x3 calibration result, the initial fitness is very low and takes a long time until if becomes meaningful. My setting is 3.5 m x 2.5 m at a concrete floor.

Lukely it converges finally.

Cheers

1 Like
2

That is interesting! Thank you for pointing that out!

Iā€™m not totally sure that the simulation code works the same as the calibration code, but they should be quite similar. Giving it a quick look through it seems like that bug is just in the simulator.

Here is the version running during calibration on the machine: