Real-Time Physics Simulation Forum

Physics authoring tools, serialization, standards and related topics • Re: Bullet serialization from Maya not saving components

2022-10-27T12:53:43+00:00

How did I solve it later, I have the same problem now

Statistics: Posted by sky10086 — Thu Oct 27, 2022 12:53 pm

Physics authoring tools, serialization, standards and related topics • How to speed up the deserialization of 'btBvhTriangleMeshShape' object.

2021-05-21T10:23:07+00:00

Hello everyone!

I am trying to take advantage of the serialization/deserialization feature to speed up the time of constructing a btBvhTriangleMeshShape object but for some reason I do not get the result I would expect.

Without using serialization/deserialization, my btBvhTriangleMeshShape object is constructed as:

Code:

//////// Inputs are://x - matrix with x-component values//y - matrix with y-component values//z - matrix with z-component values//nRows, nCols - number of rows and columns of the input x, y, and z matrixes//////// populating ground verticesconst int totalVerts = nRows * nCols;btVector3 *groundVertices = new btVector3[totalVerts];for (int i = 0; i < nRows; i++){for (int j = 0; j < nCols; j++){groundVertices[i * nCols + j].setValue( x[i][j], y[i][j], z[i][j] );}}// populating ground indicesconst int totalTriangles = 2 * (nRows - 1) * (nCols - 1);const int numTriangleIndexes = totalTriangles * 3;int *groundIndices = new int[numTriangleIndexes];int index = 0;for (int i = 0; i < nRows - 1; i++){for (int j = 0; j < nCols - 1; j++){groundIndices[index++] = i * nCols + j;groundIndices[index++] = (i + 1) * nCols + j;groundIndices[index++] = i * nCols + j + 1;groundIndices[index++] = i * nCols + j + 1;groundIndices[index++] = (i + 1) * nCols + j;groundIndices[index++] = (i + 1) * nCols + j + 1;}}// constructing 'btBvhTriangleMeshShape'int vertStride = sizeof(btVector3);int indexStride = 3 * sizeof(int);btTriangleIndexVertexArray *indexVertexArrays = new btTriangleIndexVertexArray(totalTriangles, groundIndices, indexStride, totalVerts, (btScalar*)&groundVertices[0].x(), vertStride);btBvhTriangleMeshShape *groundShape = new btBvhTriangleMeshShape(indexVertexArrays, false);

Now, I am willing to speed up the construction of the btBvhTriangleMeshShape by serializing/deserializing the btBvhTriangleMeshShape object. The following code is how I serialize it:

Code:

btDefaultSerializer serializer;serializer.startSerialization();groundShape->serializeSingleShape( &serializer );serializer.finishSerialization();FILE* ff = fopen("my_ground_shape.bullet", "wb");if ( !ff ){printf("Unable to open file\n");return;}fwrite(serializer.getBufferPointer(), serializer.getCurrentBufferSize(), 1, ff);fclose(ff);

and the deserialization code is as follows:

Code:

btBulletWorldImporter importer(NULL);if ( !importer.loadFile("my_ground_shape.bullet") ){printf("Unable to load file\n");return;}int numShape = importer.getNumCollisionShapes();if (!numShape) {printf("No terrain info found in file\n");return;}btBvhTriangleMeshShape *groundShape = (btBvhTriangleMeshShape*)importer.getCollisionShapeByIndex(0);

Let's assume that the serialization process was done offline and therefore that 'my_ground_shape.bullet' file already exists. At this stage, I would expect a speed-up when constructing btBvhTriangleMeshShape object using the desirialization approach as opposed to doing this from scratch, but unfortunately it looks like there is no gain in terms of execution time.
My usual test case has nRows = nCols = 400 and has the construction of btBvhTriangleMeshShape object being repeated 40 to 50 times. With these parameters the first snippet of code (creating the triangle mesh and allocating btBvhTriangleMeshShape) takes around 10 seconds, while the deserialization approach takes about 23.5 seconds (more than double).

I am sure I am missing some implementation point but I really can't get why the deserialization is such an inefficient method here. Can anyone explain this and give me the right implementation to speed-up the construction of btBvhTriangleMeshShape object?

Thank you!
Pierluigi Guasqui

Statistics: Posted by guasqui_act — Fri May 21, 2021 10:23 am

Physics authoring tools, serialization, standards and related topics • how to sloving adding block(urdf) crashing

2020-03-20T05:51:47+00:00

ok! when i want to add random block urdf(bring by pybullet). but when the block adding, droping in the table, they will crash and always jumping. somtimes two blocks will adhere toghter and juming down the table. so i don't konw where is the problem. and i set rhe graviety is -10.? is it problem of urdf? the collision?

Statistics: Posted by Troye — Fri Mar 20, 2020 5:51 am

Physics authoring tools, serialization, standards and related topics • Re: Bullet Physics can something like this?

2019-04-15T02:52:41+00:00

I have to experiment and check. But as far as my intuition goes it is minkowski substraction of two mesh with one mesh being rigid i.e the wheels. Bullet will return scaling factors I suppose and you will have to pass that information to your renderer and do appropriate texture manipulation. Hope this helps.

Statistics: Posted by ritzmax72 — Mon Apr 15, 2019 2:52 am

Physics authoring tools, serialization, standards and related topics • Importing and Exporting files between Bullet and Blender

2019-03-24T09:44:58+00:00

Hi everyone,

I already have a simulation in blender and I want to import this simulation into Bullet in order to retrieve contact forces between the collision shapes or rigid bodies. I also want add some features to the simulation in Bullet (e.g. implementing explosion force as body force to the simulation) which are not in blender and then export a blender file from bullet, in order to visualize the new simulation in Blender.

I wonder to know how to export a bullet file from blender and how to export a blender file from bullet?

Thank you,
Ali

Statistics: Posted by Al_Ba — Sun Mar 24, 2019 9:44 am

Physics authoring tools, serialization, standards and related topics • Why are chunk pointers updated when deserializing?

2018-02-14T06:43:04+00:00

Sorry for the rather terse title.

I’ve been playing around with the serialization/deserialization code. It’s really awesome that it can resolve pointers to structures (so long as the struct being pointed to is serializable of course), but I find one of the details of how it does this confusing.

On the serialization side, you get an id for an object in memory by passing its pointer into the serializer’s getUniqueId function, which returns an void* that’s actually hiding an int, which is weird, but that’s a separate discussion.*

And whatever object is being pointed to, whenever it serializes itself, it stores that same id in the chunk in which it’s serialized.

So far so good. All chunks have ids, and some chunks contain data members containing the id of chunks that they need once deserialization happens.

Now we get to deserialization, and these chunk ids, which are really ints, get updated to be pointers to where the chunk is being stored in memory. ???

Of course, the data members inside the chunks get updated from their int nature to be pointers (I guess that’s done through the resolvePointers function?), and then at the application level there’s the business of storing these things in maps with the key being the pointer to the chunk and the value being the pointer to the newly constructed object in memory.

So my question is, why this complex business of updating these ids? Why not just leave them as ints?

And while we’re here, thanks Erwin for this comment. It saved me at least several hours of debugging, and much suffering!

*Now that I think about it, it's probably saved as a void* because it later gets updated to a pointer, but now we're back to my original question!

Statistics: Posted by nickolai — Wed Feb 14, 2018 6:43 am

Physics authoring tools, serialization, standards and related topics • Re: How To program a physics engine

2018-01-30T00:42:23+00:00

I forgot the exact link but build new games actually published a full on guide about it. I'm sure if you Google it you'll find it.

Statistics: Posted by Ericpolik — Tue Jan 30, 2018 12:42 am

Physics authoring tools, serialization, standards and related topics • Thermal Analysis?

2017-12-14T14:10:56+00:00

Hi everyone, I was wondering if there was a way to simulate a thermal analysis experiment? To be more specific, I want to observe endothermic and exothermic transitions between aggregate states of different common metals, copper aluminium and iron. And more, depending on what funding I'll get.

Next year, I might get to do an exchange programm in Germany and do this experiment on a proper DSC PT1600. Well, I might at least if I get my essay right. Yup, I have to apply for this opportunity and I have to present a convincing essay. To write that, I would like to include a theoretical approach which will be confirmed once I'm there to do the real testing. And I was just wondering if that's possible with bullet?

Sorry, bit of a newcomer here. Looking forward to your suggestions.

Best regards,
Harry

Statistics: Posted by DirtyHarry — Thu Dec 14, 2017 2:10 pm

Physics authoring tools, serialization, standards and related topics • Re: Bullet Physics can something like this?

2017-10-24T10:57:36+00:00

The link works fine for me. Try searching the title above the link on youtube, as the result would be the same vid. Anyway, great share!

Statistics: Posted by gvx22 — Tue Oct 24, 2017 10:57 am

Physics authoring tools, serialization, standards and related topics • Can Bullet simulate gears , and elastics ?

2017-09-27T10:59:55+00:00

Hallo everybody,

I m totally new in this Forum and i couldn' t read the rules, so i offer you my apologize if i may didn t introduct myself before.

I m designing a simple machine , but i need to simulate the behavior of a system made of an elastic corp jointed to a roller chain wich is driven to the opposit direction 3 times faster than the contraction speed of the elastic corp itself.

Do you think it could be done using Bullet and 3D models from Autodesk Inventor ?

P.S.

Sorry if my english is not so good .

Thanks in Advise
NiCola

Statistics: Posted by NikoPortoriko — Wed Sep 27, 2017 10:59 am