fix cell indexing

[rubengrandia]

trick is to move the half shift (width / 2) into the rounding operation to work for both even and odd numbered maps.

We also need to correct for the fact that the center of a cell is shifted by half the resolution when viewed from the center. That is where you get the (width - 1.0) / 2 from

[rubengrandia]

Old code:

New code:

[lorenwel]

Isn't the round the problem here? I think flooring with just int i = int((x - center) / resolution); should also be correct, no?

[rubengrandia]

Seems slightly different for even numbered maps: http://cpp.sh/6c3bd But maybe I didn't fully understand what you proposed

[rubengrandia]

But you are right that it should be possible to exploit the flooring of the int, instead of doing round manually. I think this one should work: return static_cast<int>((x - center) / resolution + static_cast<float>(width) / 2);

Basically removing the half shift again and then taking the floor instead of the round. I do think it is necessary to not divide width as an integer to be correct for both even and odd sized maps.

[lorenwel]

I meant more the problem with the original implementation. This is what I meant. So

int get_x_idx_floor(float x, float center, float resolution, int width) {
    return static_cast<int>((x - center) / resolution + width*0.5);
}

[lorenwel]

Ok, we came to the same conclusion :P I don't have a strong preference which one to use, I could imagine that integer casting is slightly faster. In any case, I would replace that / 2 with a * 0.5 :D

[rubengrandia]

I tested on the robot yesterday, looks fine. Let's wait for @mktk1117 's opinion. He can also run it on some larger offline dataset to double check.

[mktk1117]

Tested from my side and it looks good.