Re: Re: Nho download paper

[Nguyen Binh]

I guess the list might be benefit from this discussion so I replied here instead.

It's true that you will need a number of LCP constraints to model linearlized Coulomb friction with each contact point. Basically, if you want to approximate friction circle by a hexagon then you need 8 friction LCP conditions with each contact.

And it's true that using this model, there will be cases when you can never get the friction force that has exactly opposite direction to the tangential velocity. Trying to re-align one friction direction according to current tangential velocity might help a bit.

And I think I need to repeat myself: 99.99% of current physics engines are NOT using accurate linearized Coulomb friction (described in papers by Stewart-Trinkle (1996) and Anitescu-Potra(1997) ). They use some sort of constraint that give friction-like behavior ( anything that resist motion can give us a feel of friction). And I am not saying that it's a bad or good thing, every model is made to serve some specific purposes. So ODE,Bullet, PhysX, Havok are great in computer games, film, etc.. but if you used them in machine desgin, robotics simulation then you should not expect the same results as in experiments. 


--------------------------------------------------
Binh Nguyen
Computer Science Department
Rensselaer Polytechnic Institute
Troy, NY, 12180
--------------------------------------------------

2009/6/23 fish <fish...@163.com>

Hi Nguyen Binh:

    I already read that method for friction model before, In the paper written by J.C .Trinkle as you recommand me to read before. I know the method is to use n friction vector in the tangent place to convex represent the friction cone as a friction polyhrdreon.But I have some doubt ,the method like the paper said for every contact point we need a matrix D belongsR(3 X N)where N is the friction vector.So in every contact point we need such a matrix ,so will it take too many memory to store those matrixs? Besides,it seems that the model can only choose one friction in the friction vectors in most of the time rather than the convex combination of them,So when the object is moving it seems impossible to calculate the friction force that is directly opposed to the velocity, this situation is methioned in Trankle 's paper "Formulations of multibody dynamics as complementary" Page .7. figure 6

So could you give me a little guild about my doubt above.

[RUoyu]

Thanks a lot ,So I think the sort of friction contraint is a little
diffenrent from the ordinary constraint(Joint,contact), because I
think it depend on the solve of the other constraint(contact Normal
force), So when solve the LCP we should take some trick for example
like first Slove the undependent force (contact),then the fricition
force, Am i Right

> 2009/6/23 fish <fishbo...@163.com>

>
>
>
> > Hi Nguyen Binh:
>
> > I already read that method for friction model before, In the paper
> > written by J.C .Trinkle as you recommand me to read before. I know the
> > method is to use n friction vector in the tangent place to convex represent
> > the friction cone as a friction polyhrdreon.But I have some doubt ,the
> > method like the paper said for every contact point we need a matrix D
> > belongsR(3 X N)where N is the friction vector.So in every contact point we
> > need such a matrix ,so will it take too many memory to store those matrixs?
> > Besides,it seems that the model can only choose one friction in the friction
> > vectors in most of the time rather than the convex combination of
> > them,So when the object is moving it seems impossible to calculate the
> > friction force that is directly opposed to the velocity, this situation is
> > methioned in Trankle 's paper "Formulations of multibody dynamics as
> > complementary" Page .7. figure 6
>
> > So could you give me a little guild about my doubt above.
>

> > <http://count.mail.163.com/redirect/footer.htm?f=http://gouwu.youdao.com>- 隐藏被引用文字 -
>
> - 显示引用的文字 -