Why does mesh with collision rise up

[MackeyK24]

When I place a capsule as a player and set the capsule right above the ground (0.1) and using moveWithCollision... Right when the scene starts... the capsule rises up a little and all the moveWithCollision the capsule is above the ground almost a half o unit...

Why does it do that... and what can I do to put my character capsule on the ground (or really close to the ground) and move around with collision ???

[Wingnut]

Hi Mackey.  I have no solution but I wanted to add a testing playground.

https://www.babylonjs-playground.com/#WWCK0#43

WASDQE to move left barrel, SHIFTED WASDQE for RIGHT barrel.  All keypresses use moveWithCollisions (in case you want to TRY TO force the barrels back to ground level - Q and shifted Q).

Both barrels (cylinders with shown ellipsoids) are being auto-moved +z with moveWithCollisions in lines 218/219.

Both are 12 units tall, and are position.y = 6.01 (just above ground).

Ground.checkCollisions is set true in line 94.  (disable == no climbing, problem gone).

Both barrels climb to a position.y (height) of approx. 12.017 units.  Related to barrel height?

Both barrels also do an unwanted sideways move, just before beginning the climbing.  Related to barrel width?

Q and shifted Q can force-move barrels back to ground level... AFTER they are no long OVER the ground plane.

I started doing some investigating... but I suck.  If I had to guess, we should be checking boundingInfo and boundingBox/boundingSphere.  Likely something with .centerWorld, .radius, .radiusWorld, similar. 

We COULD say that this is GROUND-related ONLY, but I don't think that is true.  When colliding the two barrels with EACH OTHER (using provided keypresses), I have seen similar issues.  For example, there are problems when the barrels collide from above or below one-another (turn OFF ground.checkCollisions if you do apropos tests).  Generally speaking, collisions are off-by 50% of barrel height... or 100%.... exactly those values.  That screams boundingbox/collider sizes... radius, center, extends, etc. 

But I suck, so I could be very wrong.    Generally speaking, climbing is caused by two ellipsoids in repeated collision, where the climber.ellipsoid.position.y > engine.collisionsEpsilon-amount above collideTarget.ellipsoid.position.y.  (wow, what a sentence)

I believe engine.collisionsEpsilon determines how much difference in ellipsoid position is needed... before a 'rub-around' activates.  Rub-around = climbs, dives, sideways work-arounds, etc.  Axial-aligned .ellipsoids will not rub-around, no matter the amount-of repeated moveWithCollisions() attempts.  (hold D key in THIS playground, to see it)

I may hijack collision-related portions of abstractMesh and bounding objects... into the playground... and get some console readouts.  But I KNOW we can "watch" this happening... numerically, in-console... IF we monitor the correct properties.  I'll try to determine how/where, and ALL others are welcome do to the same, of course.