I’ve been working the whole afternoon on the collision. Getting pretty frustrated now since I can’t get the algorithm to work. I found out that there are definitely more resources and examples online. Searching on the term "swept SAT" gives me some fresh results. However it doesn’t make it any clearer for me.
I did find some article that explained it and there I found you should indeed do something special when the normal is pointing into the negative direction. But it seems you should just swap the results or negate the normal when the normal is pointing in the opposite direction of the relative position direction.
To make a long story short, it still doesn’t work. I seem to get some valid values, but it’s just the combination that isn’t correct. At this point I don’t know what to do anymore. There is so much time going into this without any results. But I need an algorithm that can handle rotation and fast movement, so swept SAT is a very good solution. I already knew discrete SAT so this seemed to be a good option to try out. I also figured out most parts of the swept SAT myself as well. Though now that I looked at others explaining this method, I feel like I lost my own gathered knowledge and understand everything less instead of more.
So I don’t know what I am going to do about this situation. Maybe I can still come up with a simplified example where I can slowly build the algorithm to its final form. I just need to put this issue to rest for today, that’s for sure.