NewtonCompoundCollisionCalculateInertialMatrix

A place to discuss everything related to Newton Dynamics.

Moderators: Sascha Willems, walaber

Re: NewtonCompoundCollisionCalculateInertialMatrix

Postby Julio Jerez » Tue Oct 30, 2018 6:19 pm

I am still not getting it.
but a rigid body is a very crude model of a body.
it assumes all particles of the body are rigidly connected, but that assumption brakes down very quickly the moment a body is big.
bodies are elastic, and it is that property the let the lose energy when they collide, so on the extreme cases even a small rigid body is no completely accurate, and certainly for large bodies the force do not transfer instanstaneslly for one end to the other, oddly enough is it this property that make iteration solve perform more or less acceptable and is some case they work better than exact matrix solvers. But you can no count on that because that like say a broke clock can be right at lest twice a day, and assume that broke clock work. you can never be right for the wrong reasons.

For large bodies the inertia tensor is not longer constant and there are model the can simulate that by applying some linear and even quadratic stress.
these model still can use the rigid body model with some corrections, but the newton engine does not deal with that.

as for the question not there is not way the you can consider parts of the San to do body are movable.


in the future I will add the elastic model, but it will not use those stress model which are more appropriate to model vibrations, like building or large vehicle like ship and air plains.
instead I will use the stiff spring damper, on tretraheralized meshes, but for that we will use High Performance Computing.
that's for after newton 3.14, maybe newton 4.0
Julio Jerez
Moderator
Moderator
 
Posts: 12249
Joined: Sun Sep 14, 2003 2:18 pm
Location: Los Angeles

Re: NewtonCompoundCollisionCalculateInertialMatrix

Postby Julio Jerez » Tue Oct 30, 2018 6:26 pm

I run some test in newton about a year ago, but I put it on hold Spence there hardware was not ready for that yet, but now four cores are entry level systems.
and FEA analysis may start to be a possibility.

the good thing about the tetrahedral model is that the physics is quite uniform, but it requires lot of floats per seconds.
Julio Jerez
Moderator
Moderator
 
Posts: 12249
Joined: Sun Sep 14, 2003 2:18 pm
Location: Los Angeles

Re: NewtonCompoundCollisionCalculateInertialMatrix

Postby MeltingPlastic » Tue Oct 30, 2018 6:47 pm

Ok here is what I am going for:

I want to make a game where I can dynamically create large arbitrary objects (like asteroids and dumbbells) out of smaller convex shapes. It is safe to assume that in the idealistic game world all the particles in the object are fully rigid.

Newton so far seems perfect for this because it supports arbitrary compound collisions made out of smaller convex shapes.

Everything is working great so far - and I am so close to getting the API to where I want it.

The only big issue left is that I cannot seem to get realistic behavior from my custom objects. Things work well using the joints. But that seems more computationally expensive compared to using compounds when I want full rigidity.
MeltingPlastic
 
Posts: 237
Joined: Fri Feb 07, 2014 11:30 pm

Re: NewtonCompoundCollisionCalculateInertialMatrix

Postby Julio Jerez » Tue Oct 30, 2018 7:19 pm

everything you say there is true, using a compound is there most efficient way of doing it.
basically you want a world that is dynamics.

Code: Select all
.... this because it supports arbitrary compound collisions made out of smaller convex shapes.

you should be able you get this with compound, I do not see why not.
also the reconstruction can be increments.

by even if you solve that problem that are other much bigger once, running into numeral accuracy of linear algebra know as condition number.

I am guessing you want to make some kind of space station, but the problem is that we you make an body that big the mass matrix becomes huge, such large masses create problem when iterating with small masses.

A long time ago, there was a using that made a game "ship simulators" that has similar problem, but the had the code that simulate the ship, the want to add rigid body physics to the object in the ship.
they try to use mass and it never work, in the end the body was static and I gave the a special build that let the calculate the speed of a contact point, late that because mart of the contact callback.

I assume you can do something silicate you can have tow newton world, one for the large bodies, the space station and other large stuff.

In the world the large bodies has masses, and the have a reference to the static body in the smaller world that the use to set the velocity and angular velocity and transform matrix of each static body in the small word.

the when you step the small work the small bodies see the large body as static bodies with velocity, and the rea apropreolly.
Julio Jerez
Moderator
Moderator
 
Posts: 12249
Joined: Sun Sep 14, 2003 2:18 pm
Location: Los Angeles

Previous

Return to General Discussion

Who is online

Users browsing this forum: No registered users and 12 guests