I have a problem with behaviour of physics. I have create a dice object, with simplified collision hull. I’m dropping it from sme height and it falls on one of the edges - at certain angle. Expectation would be for dice to start rotating and jump to the side - but instead of that, it is doing elastic bounce directly up. After another bounce it stops bouncing and the starts to slowly fall on the side to finally rest on one of the sides.
Once per 20 tries or so it behaves correctly
(3rd fall, around 0:12 is what I would expect, nothing interesting for the rest of the movie, all failed attempts).
Funny part is that after switching buffer from lit to wireframe, amount of ‘proper’ simulations is a lot higher - maybe one in 2 or 3 attempts roll ok.
Why it is so inconsistent? Is there any parameter I can tweak to make it bounce properly?
In the movie above it had no physics material, but I did attach it later - and with bigger restitution it bounces more, but still completely vertical. Only when it finally stops bouncing it starts to fall to the side.
This is way too late to help the OP but the issue still persists in 4.10 so in case anyone else with a similar problem ends up here like I did…
I found by trial and error that often but not always physics actors with CCD enabled exhibit this behaviour on the first bounce. For whatever reason when CCD is enabled the first impact is registered as though the point of collision is directly in line with the object’s COG. Disabling CCD solved the issue for me but then my physics actor began falling through the landscape again which was why I enabled it in the first place. I was then able to solve my original landscape collision issue without resorting to CCD by adding p.bUseUnifiedHeightfield=0 to the end of Engine/Config/ConsoleVariables.ini (in the [Startup] section).