Julio (Newton author) on physics comparisons/COLLADA Paper

[Julio Jerez]

I see you had sent me quite few messages, this one it very interesting.

Hi Julio,
I spotted your message, thanks. Yeah, that comparison was not fair I think.
Anyway, did you have any results with COLLADA Physics?
Any feedback? I am happy to hear about problems/suggestions etc.
Is there some editor that you planned to add such .dae support?
Thanks,Erwin

What is interesting about that post is the date: Mon Dec 04, 2006 11:46 pm
This is what you said when release that when you released you paper in: April 15-18, 2007

Hi Julio,
You recent efforts of supporting COLLADA Physics in Newton are highly appreciated.
Admitted the list of physics engines was not exhaustive. If Newton's COLLADA Physics support existed before the paper was written, Newton would definately be included in the list. Note that Havok developers did contribute to some discussions in the COLLADA Physics working group.
http://www.continuousphysics.com/Bullet ... php?p=4093

It seems that over four month before paper publication, you were trying to communicate with me about Collada plug-in.
What happen did you forget, or it was not convenient to mention other technology on that paper because then you would have to make a true comparison and not the funny charade you just did to advertice your engine using Collada as platform. I call that Misrepresenations

Please stop flooding forums with your disappointment on KenB student's comparison papers. I am happy to post good things about Newton, that is why I ask you to post your new videos in a new topic. Thanks a lot,
Erwin

Flooding the forum, not at all, is this another technicality?
I do not need you to post any good things about Newton I wish you have a little more professional ethics and honesty when you write those funny papers of your where you unilaterally decided who is popular and who is not.
You started this war when you start endorsing Mr Kenb defamatory slandering of my work, and you yourself also when to so many places in the internet making insinuations and direct statements that are basically borderline lies,
about how much better Bullet is than Newton but never show it then side by side.
I think it is hip to say now that you do now know anything about Newton, guess what? Been in denial do not make it truth,
Yes Newton can be used as collision system, yes it can be profiled, Just run the demo and click show real-time profilers.
I am just responding to Pierre claims about his new rediscovery.
Quite frankly I never considered sorting a very big challenge, but he seems to be very big on that sort of stuff.
For years he seems to claiming to knows sorting better than anybody else and no one challenge his claims.
I am no saying I am better most likely I am not, but sence he made the challenge, hey why not.
All it takes is to make scenes and run they side by side: Physx, Bullet, ODE and maybe Newton. And this time let us make it public for a change.
After all you guys where the ones that came up with the "battery of test" consept designed to humiliate the lesser developers. So there is your chance hit me.

[Erin Catto]

Since this has turned into a flame war, I'll take a turn:

Julio, instead of coming to this forum with a huge chip on your shoulder, why don't you contribute some useful knowledge to the discussion?

I have used Newton a little bit and it seems to have a very precise solver. So why not share your experience instead of acting like a school yard bully?

Remember, this is about _game_ programming. It's supposed to be _fun_.

[Eternl Knight]

I had stayed out of this, because I had HOPED that the original topic would be restored. However, as Julio has decided to force the issue, I will make comment.

Julio:
Every time I read one of your posts, it is to attack someone either for a perceived slight on your engine or because they have mentioned your engine in a light not glowing with praise.

Somehow though, I fail to find any recent posts where you actually HELP others with information on thew best way to solve physics problems. For example, this thread is talking about methods of enhancing / correcting performance of broadphase collision pair culling. In none of your posts have you actually HELPED the initial poster. Instead you are still pursuing your quest to have Newton declared the "one to rule them all" (precious?).

If you want to discuss fair comparison of your engine - make another thread for it. That way people that want to talk about "Multi-SAP", can talk about it in this thread. People that want to talk about rag doll teddy bears can do it in their thread. And people that wish to discuss Newton's performance can do it in YOUR thread.

If you are not willing to add to the topic posed by the original poster please just bugger off.

--EK

Erwin: Feel free to delete this post if/when the thread is restored to discussing the initial topic. Sorry for the rant, but after a while this kind of thing gets to me.

[Julio Jerez]

Oh I see Erin you consider offensive me trying to defend my position here were all those comments and insinuation originated. Take for example these paragraphs from MR Kenb student thesis.

Page 36
Figure 4.5 shows that with ODE, the rotational energy increases in the same way as the
angular momentum. With Newton the rotational energy oscillates and slowly decreases.
Novodex keeps the rotational energy at the initial level. When observing that and the
fact that the error in angular momentum is oscillating it is likely that Novodex only
preserves the value of the angular momentum vector and not the correct values in the
vector. This will ignore the gyroscopic effect.

This has been quoted in other papers as prove of how good Novodex was and how bad Newton is.
In reality Newton was the engine that did right while Novodex prove a cheat.

In a nutshell you and I know that a spinning object with a non spherical inertia will not conserve angular momentum unless the external torque generated by the time varying matrix is added as external torque.
In fact in the absent of that compensation the body will oscillate just like in the Newton result showed, and because of the damping coefficient was not set to zero, the momentum decayed.

Now Mr Kenb Students found that Novodex conserved the magnitude of the angular momentum but not the vector, but how could that be? This can only happen for a spherical body but not for a general body. So the only conclusion is that Novodex is reading the magnitude of the angular momentum before integration, integrating and normalizing the vector after the integration if not external torque was applied. That is the only way that could have happened, because there is no known law of physic that could preserve the magnitude and not the vector. Do you have a different explanation?
If so please enlighten me and get me out of me ignorance.

I do not do that, I just let the integration go you tell me how pass the test and who failed.
These kinds of questions pop from time to time in my forum and I just tell the people what to do:

http://www.physicsengine.com/forum/viewtopic.php?t=3746

Today gyroscopic tops are not a technological brake thought as it seems to be for Mr Kenb.
I do not blame the students for been fooled by that cheap a dirty trick, what is despicable is that when they showed these bogus result to Mr KenB, he accepted them.

So that lead me to believe that he is and intellectually incompetent or that he has a different agenda. Either way I question his methods.
Gyroscopic torques seems to be a big thing in Mr Kenb Battery of test, had he asked me I could had show him the it really is a very trivial and ordinary thing, it is just that the torque is a very non linear quantity, and you would need to increate you integration step to get to be stable.
Here is how you can do it without having to claim any technological brakethought.

Code: Select all

void PhysicsApplyGravityForceAndGyroTorque (const NewtonBody* body) 
{ 
   dFloat mass; 
   dFloat Ixx; 
   dFloat Iyy; 
   dFloat Izz; 
   dVector omega; 
   dMatrix rotation; 

   // apply gravity 
   NewtonBodyGetMassMatrix (body, &mass, &Ixx, &Iyy, &Izz); 
   dVector force (0.0f, mass * GRAVITY, 0.0f); 
   NewtonBodySetForce (body, &force.m_x); 

   //apply gyroscope torque    
   NewtonBodyGetOmega(body, &omega[0]); 
   NewtonBodyGetMatrix(body, &rotation[0][0]); 

   omega = rotation.UnrotateVector(omega); 
   dVector gyro(omega.m_x * Ixx, omega.m_y * Iyy, omega.m_z * Izz, 0.0f); 
   dVector torque (rotation.RotateVector(omega * gyro)); 
   NewtonBodySetTorque(body, &torque.m_x); 
}

Now you tell me I am the bully, for trying to clean up my work, or you are the bully for housing these kind of garbage disguised as intellectual work under the title of a paper.

Julio:
Every time I read one of your posts, it is to attack someone either for a perceived slight on your engine or because they have mentioned your engine in a light not glowing with praise.

Actually that’s is the problem, you see I do not really have a problem when anybody select any technology over Newton, in fact I am the very first to recommend anyone with doubt to use something else.
My problem is with the one like you, self appointed physics engines appraisal that go around from celebrating everything not matter how immoral and corropted it is because your hatred toward me, but when confronted they are teh first how say they really never tested you make these claims just speculations.
It is one thing going from forum to forum replacing a Newton demos with a Bullet, ODE and Phsyics enticing the innocent users with source code, and I am a PS3 developer ar ex havok employee and all kind of non physics no snese, it another very different facing the facts.
I can understand people going for physx because of the soft and fliud stuff, I do not really understand Bullet and ODE, perhaps you could itemize what physics features make Bullet and ODE the third and fourth more popular physics engine. You are and expert on those I am an expert in Newton so let.
I am looking for answers to question that originated here.

[Dirk Gregorius]

Erwin, can you please stop this. Julio is totally abusing Pierre's thread here which I found an interesting contribution. I suggest moving Julio's concerns into another thread.

Julio, with all respect, how can you assume that anybody takes you serious here when you behave like a 12 year old child that was taken away his teddy bear. Maybe you can start a new thread and copy over the relevants posts and I clean up your mess in this thread here.


Regards,
-Dirk

[Julio Jerez]

I see you can not take a bloddy nose, it is tuff facing the nake truth isn'it.
In your opnion in is better to kill the message with the message when you do not like the message.
This is serious tuff, calling me names will not make newton go away, you guys started this are you just going to continue ignoring it.
I made my entry just made 20 to 30 thousand bodies and see if it runs at iterative time, if it pass that we move to a different phase.

Of course many people take me seriously, you see my physics and math is physics by the book like Sabanna, Golthain, and many others, not physics by consensus and presenations like I see around here.

[Dirk Gregorius]

First nobody wants to delete your message. I just want to move it into its own thread (just on top of this one - with its own subject that matches its content better than the current one). Second I guess a lot of people are interested to hear what you want to say, but (and now for the last time):

THIS IS PIERRE'S THREAD AND IT IS ABOUT MULTI_GRID SAP AND NOT ABOUT YOUR CONSPIRACY THEORIES ABOUT KENB, ERWIN, ERIN, ME OR THE WHOLE FORUM.

Actually the only person that seems "intellectually incompetent" is you since this is what people try to tell you now in about 10 posts and you still didn't get it.

So can we please stop this and move this to another thread?

[Julio Jerez]

It all depend what your understanding of incompetence is, if you are referring to me been incompetence for failling to understand this:

I agree that it should be tested among other various criteria, but I wouldn't consider a less physical correct physic engine the weaker one. Actually from my experience correct physic often doesn't look good or doesn't give you the expected results (of artists or designers). So a phyisc engine that creates the expected visual results can not be considered the weaker one.

Cheers,
-Dirk

I am going to grant you that, I declare myself incompetent to understand that more erroneous the engine is that stronger it is.
I was referring more to this kind of competence in correctness
http://www.terris-hill.com/products/sim ... gskill.php
This is a company that training people on oil rig exploration and they made a simulator to training real people on a very dangerous jobs. They tell me that simulator is so successful that they are licensing to schools for on the job training. That was the kind of competence I was talking about

[Dirk Gregorius]

I make it easy:

1) Do you understand that this thread was initially started by Pierre Terdiman with the topic of multi SAP?

2) Do you understand that some people might find this interesting and would appreciate if this would be discussed in this thread?

3) Why do you insist on discussing your concerns in this particular thread instead of its own?

[Julio Jerez]

Ok Mr Peirre let me see how I can respond. First let me say you do have courage I asked you and you reply answer and I really respect that.
Quite different from the Fan boys I see around here given opinions and demonizing my Character when they not like what I said.

You say that your implementation for the braadphase is not a multiresolutions grid, and as proof you give that what you insert in the list are SAP not single entities.
First what you insert in each cell do not define the algorithm what define the algorithm is that you have a array of cells wrapped by a larger cell and in each cell you have a list of objects, to go around the maintained you add bodies across boundary to each of the cell they touch.
For updates what you have is an array of fat list of self contained objects.
The fact that you have a larger grid warping smaller grids is what makes it a multiresolution grid.
The insertions of link list in each so not make the Multy SAP. You tell me where did I go wrong?

About the Collada think I was under the impression that you people designed this Collada specifications to weed out the all the lesser competitor, but now that some of them decided that maybe the only way to demonstrate that the establishment is bamboozling the people, is to do and be able to handle what they do what they do. It seems I was wrong now that Collada is not longer an insurmountable criteria to meet the condition “most popular physics engine”. As a matter of fact it seem to be a very arbitrary and suggestive, one day is console development, other days is Open source, some other is having professional support (whatever it is that, that means). What I never seen as a Criteria is how close to be an actual physic simulation the engine really is, those qualifications seems to be delegated to the lesser engines as Mr Dirk Gregorius seem, to suggest in support of Mr KenB

You said you tested Newton 1.53 and you did not see a method for queering overleaping pairs.
Top tell you the truth I did not know that that was an important function, I believe there is a function for all object s in AABB, but it has been 1 and half year since I release 1.53 and I never released any update.
I had send custom build to some user that are really using the engine for some more than demos.

Now semen you are interties in demonstrating the superiority of you new re inverted multiresolution grid seeded by fat SAP, I will be grad to send you an early version for the next released wit those special function for you pleasure.
You just have to give a description of what those functions need to do.
udword CreateBulletVolume(const AABB& box, int id);
void UpdateBulletVolume(udword handle, const AABB& box);
udword GetBulletNbPairs();
Hey maybe you can help me spot a huge bug. And Netwon can only gets better which I regeret to say will be a really bad new for some around here.

@Eternl Knight
since you just limit your comment to suppoting role of other experts, I am going the say yes to whatever you say.

[raigan2]

The weirdest thing is that you barge into a thread dealing with collision detection, complaining about your solver being excluded -- you seem to have completely missed the point of the thread. Pierre was comparing pure SAP to a hybrid approach, using some collision libraries which use SAP. He wasn't concerned with physics at all, if you had read his post you'd see that his objects were animated kinematically.

Two good reasons for his not including Newton might be: (a) it's not clear what sort of broad-phase Newton uses, (b) it's not clear from browsing the docs that Newton's collision system can be used as a stand-along collision library, without the physics. The website doesn't really mention collision as a primary feature, instead it focuses on the solver.

Or possibly he was just conducting a quick and dirty informal test, using the libraries he was familiar with (because he's worked on all 4 of them)?

Maybe the reason no one has taken a shine to Newton is that rather than share your knowledge you're keeping it to yourself? This makes your hyper-defensive attitude even harder to comprehend -- how can you feel so entitled to consideration (and outraged at a perceived lack of it) when you haven't even contributed anything?!

[btmorex]

It is one thing going from forum to forum replacing a Newton demos with a Bullet, ODE and Phsyics enticing the innocent users with source code, and I am a PS3 developer ar ex havok employee and all kind of non physics no snese, it another very different facing the facts.
I can understand people going for physx because of the soft and fliud stuff, I do not really understand Bullet and ODE, perhaps you could itemize what physics features make Bullet and ODE the third and fourth more popular physics engine.

I'm not sure what the overall point of your post is, but I think that you discount source availability far too much. For me and likely many other users, the availability of source code under an open source license is the most important feature a physics engine can have. Since Newton is closed source, that completely rules it out.

Also, you make it sound like "innocent users" are getting tricked into using other engines or something. I spent a good deal of time evaluating engines and ended up with bullet primarily because:

1) it's open source (unlike Newton)
2) it's actively developed (unlike ODE)

Probably a lot of other people reached the same conclusion as I did. Absolute correctness in uncommon setups is much less important to me.

[Antonio Martini]

the availability of the source code in a free physics engine is one of the strongest points, as you can take the parts you value most of each engine and if something is still missing you can write the rest yourself. Without mentioning multiplatform development and optimisations. Without source you would not be going anywhere.

Julio why don't you do comparative tests yourself and publish both the results and the code/data used for testing?

cheers,
Antonio

[Eternl Knight]

Julio:
Now that we're not interrupting a thread that had nothing to do with this subject - perhaps we can talk through your issues without the anger and/or conspiracy theories.

Personally, I think you are not looking at ALL the reasons one may choose another engine over yours. Let's take three well-known ones: Havok, Bullet, & ODE.

I'll summarise the issues I feel are important.

Name Recognition:
Havok & ODE have engine name recognition and Havok & Bullet has developer name recognition. One can not overestimate the ability to reassure someone financing development by simply dropping a few names. This may not be nice, fair, or even technically relevant, but it is a fact that affects developers who are not working alone and/or rely on non-technical people for finance (thats is, everyone outside the indie crowd).

Track History:
This has alot to do with "engine" name recognition, but has other elements that need to be discussed. Havok & ODE have been used in a literal tonne of games. I would say MOST of the games using ODE internally make no mention of it (due to the license). This gives these engines a "proven" status that adds comfort to both the developers (i.e. they tend to know when the bug is theirs or in the physics library) and to financers (who feel safer using a "trusted platform").

Bullet does not have this yet - a fact I fully acknowledge. I was only using Bullet for collision detection until recently due to this fact (I wanted the "convex compound" collision support). As, however, you also mentioned ODE in your posts, this detail still applies.

Support:
This is a BIG issue for those not comfortable in their knowledge/grasp of game physics. With Havok, you're paying through the nose for said support... but it IS there and it is comprehensive. For ODE, there are a number of people that have experience in it's pros, cons, and eccentricities (I know two experienced developers I can hire for this if I needed to). For Bullet, there is limited support from Erwin himself, but there are also three others on the forums that seem to know their way around the innards quite well too. Everyone I have spoken to have been polite & helpful (even if they did not know the answer to the question I was asking).

With Newton, there is only you. And, being frank about it, from what I have seen/heard from your posts - I doubt the courtesy I would recieve if I were to approach you with something that you disagree with.

Code Mixing/Hacking:
With Bullet & ODE, I can mix and match my methods of collision detection, constraint solving, and so on. I can specify the maximum number of contacts, use (or not use) persistent contact point manifolds, etc. I can mix the two engines together to get whatever I feel is the best combination of their strengths (e.g. Bullet's collision detection & contact manifolds with ODE's LCP solver & constraints). I can experiment with a variety of techniques to determine (for a recent example) the best broadphase method to be used for different game scenarios. This is all possible because I have the code.

Newton is closed in this regard. I cannot tweak the broadphase depending on the game scene specifics. I cannot hack out bits of it I do not need. It has a decent API, that I am not arguing, but it is a static API with changes to it only possible by yourself. Never overestimate the desire for a developer to slice away and/or hack into parts of a library he thinks can be optimised or discarded.

Updated Development:
Simply put, Havok & Bullet are frequently updated. ODE has pretty much died (there is some smouldering life still there, but I am not expecting anything major to come from it). Newton's last release was a year ago. Enough said really.

Accuracy vs "Good Enough":
This doesn't take much thought really. If we wanted the best accuracy in our simulations - we would not be looking at Bullet, ODE, Havok, etc. We would be looking at the engines used in manufacturing development, industrial robotics, and so on. Like computer game graphics, it does not need to be a perfect simulation - so long as it is "good enough", the developers and end-users will be happy.

Summary:
On the whole, I think you are taking offense at the wrong things for the wrong reasons. If you have a need to PROVE yourself and your engine against Bullet, ODE, etc. Write your own comparison code. You have an issue with the way someone else represented your engine and/or conducted their tests - write your own and publish it.

Put the code and the results up on your website. Ranting and fuming at others for not doing it the way YOU want it done is obviously not getting the results you want (unless flaming and trolling IS the reason for your posts here). Put a day or two's effort into it and PROVE yourself right rather than just posturing that if someone else did things your way Newton would prove superior.

--EK

[Julio Jerez]

I am not asking for any consideration from, quite frankly I do not really need it from anyone here.
This is not a post against anyone here this is about two immoral individuals: Mr KenB and Erwin Coumans.
I am asking for a clarification from Mr KenB battery of test which Mr Command decided to showcase as a TROPHIES of Newton impopularity.
http://ode.org/pipermail/ode/2006-December/021186.html


I am also asking how was it is that Coumman arrived to that conclusion 7 month ago that of all of the physics solutions out there: Tokamak, TrueAxis, Open tissue, Dynamek, Newton and so many others, he arrived to the conclusion that his is the fouth mot the third more popular.
I just want and public answer to that other that changing the title of the paper, implication of his employer, saying that consoles what Make his library popular etc, etc, etc, tec.

To the Fan boys i am not go to say anything. I will respond to Mr AntonioMartini thought.

Mr AntonioMartinim, I actually respect your convictions about the importance of source code. It is a shame you fill absolutely not respect for mines. Few months ago a person was asking for a comparison of physics solution and he ported the same question in few forums. Anyway this is a link to the question in ODE:
http://ode.org/pipermail/ode/2006-December/021184.html
And this is what you replied:
http://ode.org/pipermail/ode/2006-December/021187.html

I can see that you, like Mr Kenb, have a great deal of confusion with Gyroscopic forces stuff,
I am going to recommend to you to read my post about how to reproduce, http://www.physicsengine.com/forum/viewtopic.php?t=3746
It really isn’t that difficult Mr AntonioMartinim and certaintly is not something that can only be done with open source engines.
It appears that your ignorance about such trivial an ordinary physicals phenomenon compelled you to assume that ODE was the only software capable of doing that, when in fact I believe ODE is notoriously famous for not been capable of doing that.
You also posted this as example of exelence in robotic simulation http://www.droidlogic.com/

Now since you posted a link to what is your standard of robotic Excellency, let me show you what the expectation of Newton performance is:
http://www.simudrive.co.nz/
It is a commercial training simulator for given licenses to drive pooler at the airport. This is a video for the actual presentation to with the actual instructors at the airport
http://video.google.com/videoplay?docid ... 7815823646
Also field hmore closet;y related to robotic here are some real robots that use Newton for simulation and are expected to math 100% the real prototypes
http://www.pobot.org/-Robot-Race-.html

As you can see Mr AntonioMartinim it all depend what level of expectation the user have for the technology, it seems that in the open source crowd it more like a cult crowd, and expectation of cuality is quite low, I do not have that luxury and I really have to deliver a working product. Like a fan boy here just said Netwon was last release was year ago, actually it was a year and a half.
What the bright defender do not really undeterred is the laws of physics do not get old with time, Newton do not really needs updates it is a newtonia rigid body simulator it has all teh feature it needs, so is a fresh today as it was them. I must thanks Mr Eternl, than Eternl I kwn yuo are a the the beleiver for the cause of open source, BTW hows that Mass ratio problem you were having? hows that work out for you these days, I know, I know teh powere of the souce code will allowe form soem magic tweks and badabing badadum will be "good enougth.

Mr AntonioMartinim, this is not the first time you and I have differences of opinions, you seem to have a great confusion about the Barraf LCP implementation of Danzig algorithm.
You blindly promoted the method as and O(n3) time complexity and I have to prove to you that it is a consistent O(n4), Perhaps if you read more actual literature you will be more familiar with these trivialities, maybe: Linear Programming And Extension By Danzig himself, or maybe The linear Complementarity Problem By Cottle Pang Stone, can clarify you from soem confusion so that you at least give more informed recommendations.