I want an concept, please :). I’ve applied transferring AABB -> AABB collision detection (3D), returning the time of affect.

When I calculate time_of_impact * velocity I generally have accuracy issues, e.g. participant ought to keep at y=64, however after multiplication it is y=63.999996. So the following iteration returns time_of_impact < 0.0, however its decision, would not clear up the collision, as a result of once more it stays at y=63.999996. Moving the participant ahead then collides with the ground tile from the facet (x or z axis).

How can an algorithm deal with numeric inaccuracy?
I take advantage of 32bit and did not check 64bit precision, however in some unspecified time in the future, it also needs to occur.

Thanks up to now.