Joints stiffness

[JoeJ]

It is in todays version with the added Skeleton Solver Mode Interface,
and in yesterdays version when i complaned gainst the missing stiffness.
I guess i missed 1 - 2 versions before that.
The working version i have has folder creation date 17.08.2015.

[Julio Jerez]

Ha good I will check that out, I played with the box contact collider, maybe I add the bug at that time.
I was to make a sample test to reproduce this in the sandbox? maybe a box with a strong force push it down would do it?

[Julio Jerez]

on a side note, Google code finally close it for goo, so I can no use as the dev system.

but on the bright side the new GUI version of Github has a label mark each time you commit so it should be eassy to get older version now.

what I do not like of Github is that it try to do eh same nonsence the system like perforce do, the link the directory to the archive so you cannot sync to any director only less you do a full clone to a folder an the created a new entry. with SVN torstoi you just make a folder and click svn check out.
anyway, this is what made me work locally in google code, and the commit to github, but that I can't do no more. That way I can have several feature mini project going on at the same time.

[JoeJ]

I was to make a sample test to reproduce this in the sandbox? maybe a box with a strong force push it down would do it?

I guess that should do it, maybe oszillatint torque too.
Use a animated kinematic platform body - i'm not sure if it happens on dynamic bodies too.
The problem does not show up if there is none or very slow animation on the platform.

[Julio Jerez]

let me see if I get this
you have a kinematic platform the you are moving and setting velocity, then if I place a body on top will penetrate the body. and this is new since the late few commit.
this give me the change to make a kinematic demo with moving body.

[JoeJ]

Yes.
I just removed the ragdoll and dropped a simple small cube above the moving platform.
Drop hight was small but the penetration is quite large - so i think it's easy to reproduce.

[Julio Jerez]

thanks I try that tonight, this take precedent to everything we can no have this kind of bugs.

[Julio Jerez]

Ok Joe, I think the found the broke the box collision, at least was the part that made the box collision soft. but we may be talking of tow different box but I am almost sure is the same problem this was the problem.

also I added the LCP functionality to eh skeletal solver, it is the ietetaion by I believe is one pass is no enough because it jitter a little in you 20 link demo.
I believe this is because the that sample is actually quite demanding sin all the motor are in fact aligned so it has a hard time find the optimal break, I will experiment with more passes later,

in theory the normal passes in equal to the number motor rows, for the exact solution,
so the demo will have to do 60 passes. however by use clever direction the can be reduce a grate deal. The demo is doing only one pass and the structure is straight as an arrow, but with eh that jitter. Please try it again see if the is better.
you have to enable the option by doing this

Code: Select all

    NewtonSkeletonContainer* const skeleton = NewtonSkeletonContainerCreate (scene->GetNewton(), parent, NULL);
   NewtonSkeletonSetSolverMode(skeleton, 1);

[Julio Jerez]

Oh Joe I just realized that I did no checked in the solve fix, I only check in the box collision box fix.
I will check the second part this afternoon.

[JoeJ]

Ok (i, was wondering SetSolverMode did not make a difference)

My box collision bug however is still there.
You want me to upload my older but ok newton version somehere?

[Julio Jerez]

The bug I fixed take care of the eassy penetration when a box was already in penetration.
I did that few days ago an made a mistake.
This is a different bug that I probably added a while back probably when I added the aggregate or the persistent broad phase, it will be hard to go back that far to pinpoint when it happened.

yes please could you upload that working version some where so that I can compared it.

[JoeJ]

http://we.tl/XIXLmdXIcG

This is most actual version without box bug.

Also i have modified the Joints demo with a flexible point to point in my joint.
It worked like this for my ragdoll but not so well for the 20 bodies snake.
You may wanna copy this for future tests.

[Julio Jerez]

wow this is weird, I was sure I checked the skeleton solve fix, but when I look at github from work it was not there, but from home it is.
I added a dummy check in, can you see If the code is there? the comment in the check in history is
"added LCP functionality to skeleton solver" should be the second form the top in github

[Julio Jerez]

also I compared the archived version to the current one and the it seems the bug is the same function that I fixed. It is this function here

Code: Select all

dgInt32 dgCollisionBox::CalculatePlaneIntersection (const dgVector& normal, const dgVector& point, dgVector* const contactsOut, dgFloat32 normalSign) const

I think you still have the older version that was there befopre I commited the change, maybe it was a delay in Github. Can you please sync and check that you have the latest and try again?
both the ragdoll and the collision should work near perfect.

if the collision bug is still there. Can you replace that function I listed with the one in your good version so that we can rule that our?

note: I also added your latest modification, the point to point flexible demo, and even that one seems to work fine, of course I do not know how it suppose to do. but seems fine.

the github label is "Added new Joes flexible point to point demo."

[JoeJ]

Looks perfect now, congrats!
I can't find anything to complain against using my ragdoll.
Really, really awesome work!

I can't see the jitter you mentioned on the 20 bodies snake - what do i need to do to provoke it?


I think you're right about the delay - meantime there's "added new Joes flexible point to point demo",
So i might not have the most recent version again, and i guess there is an additional lag when creating the zip file that i use.

However, the box bug is fixed too.