Comments on: Separating Axis Theorem (SAT) Explanation

By: meming

meming — Tue, 06 Jul 2010 22:10:33 +0000

I totally understand whats going on, its just I dont know how to put this in code, I give this tut a 90% ( which is great ) but if you put a little code under each step, (especially when we create the “axis”) this would get an instant 110%. You wont have to reply to this, ill just spend an hour on google, but future viewers will want the same.

By: Adam Baker

Adam Baker — Thu, 20 May 2010 06:55:17 +0000

In fact, now I have some feedback: it would be worth emphasizing more that it’s necessary to use normals from *both* polygons as axes.

By: Adam Baker

Adam Baker — Wed, 19 May 2010 06:49:03 +0000

This is a great explanation, thanks!

By: Andrew Sevenson

Andrew Sevenson — Sat, 27 Mar 2010 11:05:24 +0000

Hmm – good question. Its been a while since I wrote it so I had to have a look at the code again.
At a quick glance it looks like the vOffset is the vector between two shapes, and the sOffset seems to be a scalar used to calc the projected points of the second shape. I can’t remember exactly why it is there – my instinct is that it would be to keep the objects in the same coordinate space – but if yours works without it, maybe it isn’t needed :p

By: wwwebber

wwwebber — Wed, 24 Mar 2010 09:43:30 +0000

Great post…question, I used your code as a basis for my own SAT test..question though–what is the purpose of sOffset and vOffset? Intersection tests in my version, work perfectly, without using either of these.

By: Andrew Sevenson

Andrew Sevenson — Sun, 31 Jan 2010 11:51:01 +0000

Ah yes – you are absolutely correct. I’ve fixed that now – thanks for that

By: Vanderful

Vanderful — Sun, 31 Jan 2010 11:38:36 +0000

Great tuto !

Just some details:
In your first sentence you talk about concave polygons instead of convex ones.
The axis to be kept is the one which has the minimal overlap (not the greatest).

Anyway, really nice work !

By: Andrew Sevenson

Andrew Sevenson — Sun, 31 Jan 2010 01:51:12 +0000

Hi Will,

I had a quick look at the implementation you posted. It appears to be checking for a collision only, and not actually figuring out how to separate them. If you were to use the ‘axis’ property directly after that method you would always get the last axis that was tested, which is not necessarily the the shortest path of separation.

To get the right axis, you would have to keep track of which axis had the shortest amount of overlap. This should be as simple as having another vector variable declared and checking it each time an axis is tested. If you get to the end and there is a collision, simply use the vector you have stored for the separation.

At least, that is how my code works.

Hope that helps

By: Will

Will — Sat, 30 Jan 2010 06:09:59 +0000

The problem seems to be that the axis given for separation is the wrong one, any help?

By: Will

Will — Sat, 30 Jan 2010 01:22:29 +0000

I’m writing an implementation of this, and I can’t figure out how to find the vector that separates the shapes (called collisionInfo.separation in your implementation). Having just read through your version, trying to write your way of finding it, and watching my two shapes fly right through each other, I was wondering if you could provide some light on this topic.

BTW the original implementation that I based mine upon could be found here (if you could put how it works in the wiki, that’d be awesome )