Reduze max_z and increase min_z (f.i. max_z = 1; and min_z = -1; ).

You can use ent_buffers to write your own bounding box c_trace and movement, it's not that hard for sidescrollers and gives you good practice.

For sidescroller I can use simple two short traces, from each corner. That will also work great. I wanted this to work with FPS too. Thank you for an idea with flat disk, I'll give it a try. BTW, what about problem I have with collisions between rigid and character (see pictures)? JCL, there is a small misstake in script (which you gave link to):
It misses ">" at the end.

If you get no collisions, you have a wrong setup. For instance, you move an object that is neither kinematic nor a character. Collisions are supported for character controllers and kinematic objects only. Other possibilities are that your object penetrates the ground, or that you've set some strange physics flags.

Just begin with the very simple example from the link that I posted. It's only a few lines of code and shows you all 4 movement methods. Also check which version you're using - pxent_moveglobal for chars is only supported in 8.20.

Yeah, as you can see in my example, it's very simple. I got collusions, but they are strange, models penetrate in each other. See pictures for that. I tried CCD as well, but I get same results. BTW I got same result with your example too. I checked my version and it's 8.20 (the last one as I know).

I lowered "my.max_z" to "1" and changed "my.min_z" to "-1" as well, but I have same problem with distance to ground as before.
Why don't you make additional BOX hull for "c_trace"? Why do users of PhysX think about workaround for ellipsoid trace?!

BTW here is a video, to show you my problem:
Youtube link

You're maybe confusing the c_move collisions with physics collisions? Lowering my.max_z and my.min_z has no effect on physics collisions when an entity is already registered. It has only effect on c_move, c_trace and so on.

No, I did that right before setting it's "pXent_settype", but I have no idea why it doesn't work.
BTW, as I already asked, why don't add BOX hull for "c_trace"?! Why we need to think about workaround?
Or make another gravity forces for CHARACTER object! Ellipsoid hull has nothing to do with BOX hull of PhysX.