Several questions about Newton functionality

[FSA]

Hi! I have some questions about NewtonPhysics.
1.) The demo "SimpleSoftBody" crashes. Are you still working on that?
2.) In the demo with destructible geometry "StructuredConvexFracture" I can't to anything. Still in progress?
3.) Is newton going to support water and smoke simulation in the future? Like in this video: http://www.youtube.com/watch?v=1C6LrDzjfRw

Tank you for answer.

[Julio Jerez]

"SimpleSoftBody" and "StructuredConvexFracture"
I start then but there is never demand for any of the stuff.
so focus on what people are using most.
if you are intested on those funtionality I can complete both demos. (one at a time)

on the youtuve video, that demo is a nvide scan. basically that flid simulation is a brute force simulation usin a solid N x M x R cube of
a velocity field, any one can make thos kind of demos,
the problem with that kind of demso is that if I make one ten peopel will try to use it and at some point they wil relize taht is simpel no practical.
The simulation using Particles hydrdynamics is a posible solution, but still onel apply for small confinement.
that's a possiple route that I can use for small body of fliud, and for particles.

are you planning to use any of that funtionality?

I ask because that take lot of work, and most offetr people ask for functionality and after a week or tow of work then they move on to something else and I never get a feeback unlt soem oen else come alone.
you have to relize that I am only one person, and I have to select carefully what I am going to spend time on.

If you wnat I can resume teh sif body this weekend.

[FSA]

I know that work. I'm one programer too who creates a 3D-Engine
SoftBodies and smoke are the most important for me. StructuredConvexFracture will only be feature in my engine but is not planned in our game. If you can complete the softbodies that would be good. Because then I can start to implement NewtonMesh to my engine and this is going to take a lot of work. I have only one request: If your doing the soft bodie demo, can you make it as simple as possible? Because this functionality is very important to me and the destructible geometry code scared the * out of me
And don't worry: I'm using Newton since version 1.53 and I will also use it in the future. So if you make a feature, it will be implemented in my engine

[Julio Jerez]

Ok I will resume the softbody this weekend, let use see how far I can go.
if I remember the almols all teh funtionality was there, I had to write the voxel collision for it.

It si probably brokend now with all the changes I made, bu I will try to bring up to date,
even with flat collision and then afet we have somthong running I work on eh collsion.

[pHySiQuE]

I'm interested in soft bodies for cloth and vegetation, but won't be able to pursue it for another 4-5 months.

[JoeJ]

There was a cool video recently showing physx can handle cloth, fluid and rigid stuff interacting seamlessly with each other (cloth balloons filled with water...).
This is awesome, but i'm sure it still requires to much memory and GPU power to be practical. It's better to spend this elsewhere, like GI and 'more than stacking crates' physics.

I'd like to have soft bodies / cloth in newton for things like: A soft sofa, a soft huge monster, a clothy ceiling above a main entrance where a character can fall into, etc... Meaning i'd be happy with a detail level doable on CPU. But even that is bottom wish list.
Vegetation is a really good example for actual needs.

High detailed character cloth or smoke is GPU stuff and may not require interaction with rigid bodies.
I do not expect this stuff in any physics lib - easy to do but related very specific to custom needs.
Maybe this stuff should wait until GPU and CPU are one thing and a good general solution is possible.

[Julio Jerez]

Ok that settle it, this week I will resume the soft bodies. I have neglected for too long.


on the youtuve video, that demo is an NVidia showoff. basically that fluid simulation is a brute force simulation using a solid N x M x R velocity field cube,
any one with basic knowledge Stoke and Navier fluid equation and do those kind of demos, there are many made by Baraff and other people for more than 20 years now.
The one this wit that is the Particles rendering and that it is using about 1000 indivual Cores. believe me that demo is 100% impractical.
Nvidia want you to buy a PC capable of using tow or more GPU, so that you can use one GPU car to render that demo.
The reason those demo do no scale is because fluids are volumetric solid that are compact, they grow with a the power whit GPU grow exponentially
for that kind of fluid simulation to become in real time standard in graphic, the GPU would have to go for a least 10 generations.

The problem is that GPU are reaching the limit of parallelism, so no I do no believe that volumetric fluid simulation will even become main stream.

The are some crackpot out there that think they can make it happen using tricks and inventing brand new laws of physics,
I have to tell you, Nvidia spend many millions of dollar at year providing programing support to high end developer integration the particles stuff in the games and the result always leaves a lot to be desired.

The reason is this. When they approach the developer, they wish the developer make the game only high end users, but the developer has different priority, the developer wants to cover the middle and lower end user,
they want to sale Games while nvidea wants to sale cards. In the end no matter how much the Nvidea engineers hack a game to add particles or fluids, the developer will always tone it down to the minimum possible.
The result is that you see youtube videos of a games using Nvidia physics, but it is running on a 10 or 15 thousand dollar machine that I doubt you will find 10 or 15 people owning a rig like that.

[Bird]

"SimpleSoftBody" and "StructuredConvexFracture"
I start then but there is never demand for any of the stuff.
so focus on what people are using most.
if you are intested on those funtionality I can complete both demos. (one at a time)

I'd very much like to have those features too!. If memory serves me you were working on the Boolean operations in order to do dynamic fracturing but weren't able to implement them to your satisfaction at the time.

-Bird

[Julio Jerez]

I have all of the core foundation for the destruction, I realize that I do no nee a full Boolean all I need if a semi Boolean clipper, basically a function that can clip and arbitrary mesh again a convex mesh. I wrote already.
what is next Is presentation. I move to the editor to write a basics authoring. because it is quiet hard to make a demo procedurally.

her is were there is the dilemma, a too simple demo does not look convicting and a complex demo and people complaint that is too complex.
you can see that the very simple demo that does the rock destruction is almost the simplex thing that can be made, and people find too complicated.
Even the newton Mesh people find to complex to use.

[pHySiQuE]

The boolean destruction stuff is pretty interesting to me, since we already have geometry from our CSG map editor that is always made of convex hulls.

However, the main limitation of these things seems to be structural breakage dependence...If you have a column holding up the ceiling, and you break the column, you would naturally want the ceiling to crack and fall down. I remember seeing a demo of something like this, and you were just blasting chunks of concrete off a steel framework. So the structure itself never can collapse, you can only break chunks away from it. This makes the applications for its usage somewhat limited, so I am not sure how well I could put it to use.

[Julio Jerez]

Oh not, you are mistaken there, I called structural destrution beacause it will destruction correctly just like you describe.
it is actually very eassy to do with a graph and using a coloring algorithm.
also geomery that is already made out of convex peoeice offer and simplification step and soeul eb moch easies to get it going.

I will make the demo after the soft body.

[JoeJ]

Julio, have you thought about simulating fluid by linked volume polyhedra? Example is water in a glass, shaped like a top cut sphere,
and if someone makes a hole in the glass a cylindrical downwards flow happens, subtracting volume from the glass water and building another volume on the floor.
The water tries to change its shape by forces and also tries to compensate volume. But if all Water is on floor and becomes inpractical large and flat surface or if it should explode to too much small pieces it is allowed to vanish...

Hard geometrical work, but destruction / boolean stuff is similar.

[FSA]

By the way: It would be very useful for me if you could add the

Code: Select all

void NewtonBodyRemoveSimulation(body)
void NewtonBodyAddToSimulation(body)

functionlaity first. Some pieces of my project needs them. Keep me up to date

[elvencode]

Just my 2 cents about important functionalities:

- of course the priority would be to have the core functions (collisions, joints, etc) working in every (most?) situations. Yeah, now it's alpha and introducing new features break things here and there but every n versions there should be a revision with those base things ok (maybe tested by us to confirm everything?)

- personally i'd like to see some streaming functionalities (if not present, i'm not really that deep in using the engine). In large scenes i'd like to split the enviroment in smaller sections and load them when the player visit them (like in an open world game). This could be useful so that the developer can cull out part of the scene at the begin without having lots of things in the physics engine and having the culling done entirely inside of this last one. Also, because i've potentially quite a large scene i divided the world in small sections in my game. Every section (like an octree cube) has an offset value from the origin and of course limited set of float numbers in it (ex. i've sectors large 1000 in each side, inside that objects x, y, z coordinates of the positions are between 0 1000 or -500 500, the sector at the origin has offset 0,0,0 the next one you add the side length, etc). This way my render engine doesn't have to manage problems with high floating point values. Don't know if there are the same problems with the physics engine. I read this in a feature proposal for Ogre3D and decided to use the idea in my software. If you want a large set of collision data you can download some here http://kayin.pyoko.org/?p=2218 (there's a link at the end of the article) It's the collision data from a game called Dark Souls, a third person RPG with a large world all connected together. The files are in OBJ format so it should be easy to convert it. The data is the collision one so it should already prepared in a good way (no poly soup of bad vertices). In a world large like that i think they stream the collision data somehow, isn't it? Each file is a level of the game but ingame they are joined together when you go around. Take it more like a stress test data In game they walk around, have ragdolls, smash things with pieces going around, round balls rolling on ramps and throw things like arrows around so it's not only the walkable are

- about cloth i think it's not important to have very detailed surfaces (that probably need some GPU simulation, maybe an idea of a future OpenCL implementation) but "normal" features for games. A patch (maybe not necessarily a rectangle) with some anchor points, stiffness, wind simulation, collision and possibly (but not so important) breaking limits. Typical examples: flags (realtime and also, if possible, just simulating it offline for some second and record the frames of the simulation. Like simulating a flag with some wind and then just render the flag with the recorded data overtime without having newton simulating it during the game. Could be useful for lower end machines), patches anchored to a animated character (scarf, short cloak, long beards, stripes that simulate fur that move with some stifness on an animal monster when moves) or rigid body (a wood cart with a cloth covering it, a rectangle set of nodes inside the cloth is anchored to the borders of the cart so when you move the cart the cloth dropping down just wobble a little) or static enviroment (spider webs that when the character touches them in a cave they stretch than break. don't know if grass can be simulated with cloth simulations). Another cool thing could be burning simulation: you decide when a fire starts in the grid of the cloth and the fire spreads through the nodes and break them after a certain time. The render is notified when a node starts burn and when break so that it will render the fire correctly Probably it can be done by the game developer but could be funny to have it somewhere in the future in the physics engine

- about softbodies i think that it's not something used much in games other than maybe physics puzzle ones. when i see some soft thing probably it's an effect of ragdoll (in Demon's Souls you have slugs and another serpent like creature that when die the corpse wobble around like if it was soft) or something preanimated (like a piece of a enemy that moves like it is soft but with not interaction with the physics). Maybe if you have antennaes in a character and you want that they wobble when it turns around you need soft body simulations isn't it?

- fluid simulation: i agree that complex simulations are not feasible with complex enviroments like the ones seen in most games. Maybe they will do something in confined places for puzzle games but fully fluid simulation i think is both not simple nor useful, just cool to see. Maybe a simulation like in Portal 2 where you just show some water shot from here to there with collision is enough for some gameplay or special effect (fountain, waterfall?). Are they using metaballs in Portal 2?

- particle simulation: i'd like some more of this. Yeah, it's not so difficult to do a particle simulation in game (well maybe optimizing it a little more) but having it interact with the enviroment is another thing i mean, simulating for example some sparks bouncing around or some dust that whirls or moves around from a wind force and touches things in the meanwhile. i think a simulation of < 500 particles is more than enough for most of the situations, would it be possible without the GPU? Just having every particle systen colliding with static and dynamic objects but not between different particle systems.

- wind simulation: the points explained before but implemented in a general way so that potentially i can set a tornado force in this place in the level then if i have a box that is placed there it whirls, or a particle system (smoke) or a flag. Of course with possible filtering (saying "not apply wind of this object but for this yes). Having a simulation of various types of wind, from directional to circular (and others if there is literature).

- fracture simulation: it's cool but somehow one can now do in other ways (by breaking joints in complex objects, by replacing an object with small pieces when smashed). But if working i'd implement something like that

Just some ideas for the future. Of course the first one and maybe second can be important, everything else it's pushing the level for cool things to show in game

[FSA]

You could use

Code: Select all

void NewtonBodyRemoveSimulation(body)
void NewtonBodyAddToSimulation(body)

to stream your world. But Newton doesn't eat much memory. So culling physics is not important (my opinion). You have AABB check also before the collision algorithms I think.

SoftBodies and Co: It seems that you're not very familiar with NextGenration engine development