New physics engine in academia

Please don't post Bullet support questions here, use the above forums instead.
Post Reply
ngbinh
Posts: 117
Joined: Fri Aug 12, 2005 3:47 pm
Location: Newyork, USA

New physics engine in academia

Post by ngbinh »

FYI: a new physics engines implemented by academic researchers:

http://www.cs.rpi.edu/~sberard/dvc/

The project is guided by Jeff Trinkle and daVinci Group (Jeff Trinkle, David Stewart (ST timestepper), Jong Shi Pang (LCP), Vijay Kurmar(locally complaint model),etc..).

Thanks.
Etherton
Posts: 34
Joined: Thu Aug 18, 2005 6:27 pm

Post by Etherton »

It seem physics engines are dozen for dime now. Here is another one

http://www.ThePhysicsEngine.com

They both promise in the future they will be better than the others.
Erin Catto
Posts: 324
Joined: Fri Jul 01, 2005 5:29 am
Location: Irvine
Contact:

Post by Erin Catto »

But just think of how many rendering engines are out there.

More is better, especially when they are open tech/source.
Eternl Knight
Posts: 58
Joined: Sun Jan 22, 2006 4:31 am

Post by Eternl Knight »

Yeah, but there are only a handful of GOOD rendering engines out there. Most the others are... "underdeveloped". This is how I feel about most the "new" physics engines that come out lately.

I'm hoping for better from the DaVinci one as it progresses, but I feel that "perhaps" it's going to remain the way it is for the forseeable future. That, and the third party components prevent it from being "easily" pluggable into game engines (at the very least).

--EK
ngbinh
Posts: 117
Joined: Fri Aug 12, 2005 3:47 pm
Location: Newyork, USA

Post by ngbinh »

You are right!
Most academic projects suffer from long delay and kind of "indefinite" milestone. I hope that DaVinci will be an exception.

About the third party currently held in this version, in my new design, everything will be implemented as components, "default" components and 3rd party components. So if you use default components, there will be no dependency(of course, except core libraries like STL,OpenGL,etc..)
Currently I only see the difficulties in LCP solver and collision detection. We will definitely have a specialized LCP/NCP solver but it may not be implemented in the near future. And we don't have any plan to write our own collision detection, except a super quadric collision detection package under development. Bullet will be one of collision component.
The other possible high level dependency could be COLLADA. I plan to use COLLADA as content creation component, but it won't be a requirement.
Eternl Knight
Posts: 58
Joined: Sun Jan 22, 2006 4:31 am

Post by Eternl Knight »

I was not implying that DaVinci was a "flash in the pan" type project. I am personally hoping it isn't, as a complete engine that optimised for accuracy (as compared to tose that optimise for speed) would be very useful in a wide variety of simulation scenarios. Not everyone wants/needs something that will sacrifice realism for speed & "aesthetically pleasing" simulations. I would gladly allow an engine churn for an extra second per frame in some circumstances.

Besides the fact that I would love to see an NCP solver implemented for physics simulation :)

I simply had a few alarm bells go off due to the fact that (a) it is using third-party components I cannot move to other platforms, & (b) it is a 1.0 release but is relatively limited compared to most engines I have played with.

--EK

EDIT: Spelling mistake
Last edited by Eternl Knight on Thu Feb 16, 2006 9:12 am, edited 1 time in total.
Etherton
Posts: 34
Joined: Thu Aug 18, 2005 6:27 pm

Post by Etherton »

Forgive my ignorance but I do not know what an NLP and NCP solver is.
My impression was that the LCP condition for constrained dynamics comes from the formulation of the equation of unilateral constraints that lead to a Linear Complementary Problem.
What kind of formulation leads to these NCP or NCP and what are those?
ngbinh
Posts: 117
Joined: Fri Aug 12, 2005 3:47 pm
Location: Newyork, USA

Post by ngbinh »

NCP = Nonlinear complementarity problem.
You will have to use NCP when:
1) Your big matrix M (inertia, constraint jacobian) is non-linear. This thing will arise if you use 2nd ordered time stepper(trapezoidal,...)
2) When you use non-linear constraint. This is quite a must if you have smooth geometry (super quadrics), or when you don't want to model collision plane as infinite half-plane (which is causing so many trouble). Our Matlab code show that with this non-linear constraint, we won't have any problem with sharp object.
Remotion
Posts: 22
Joined: Sat Sep 24, 2005 10:01 pm

Post by Remotion »

NCP this seems to be interesting!
Is the Matlab code available, to try to understand how it work?
ngbinh
Posts: 117
Joined: Fri Aug 12, 2005 3:47 pm
Location: Newyork, USA

Post by ngbinh »

Not at this moment!
We need to clean those things before releasing them.

About Nonlinear stuff, take a look at: James Anthony Tzitzouris PhD Thesis
http://www.mts.jhu.edu/~jimt2/

Or a shorted version: a technical report here:
http://www.cs.rpi.edu/~trink/Papers/TTPtr.pdf
Post Reply