ESA Summer of Code Project Proposal
Non-vacuum gravity simulator for the BRL-CAD Solid Modeling tool
- Name: Oana Niculaescu
- Mailing List ID: firstname.lastname@example.org
- IRC ID: elf11
I am a student at the Polytechnic University of Bucharest, Romania, in the second year at the Computer Science faculty. I have a good background in programming and I had an internship with a local company here in Bucharest on game design and iOS. I also have a little bit of experience with CUDA and GPU programming. I am currently available for working 40 hours a week on this project. More details about my background are provided by my CV and I will link you my Github account, where some of my work is presented.
Brief project summary
The current non-vacuum gravity simulation available in BRL-CAD has not been developed to its fully potential and it is limited. My proposal is about continuing the work previously Abhijit Nandy did on the project and improve on his results. Since the Bullet Physics engine has been previously used and for this project and there are integrated tools for BRL-CAD, like gqa and rtcheck, that perform exact collision detection I consider there is no further need to switch to the Open Dynamics Engine (ODE). Basically the scene is read from a list of objects, the Bullet creates it and then forces are applied such that the position of the object will be updated, then the two BRL-CAD integrated tools for collision detection will be called and the collisions will be solved. Previously aligned axis bounding boxes have been used in collision detection and I consider it is a good thing to have a continuation with those.
Detailed project summary
Since the initial developer for this project has been intending on implementing a command in the Archer GUI for solving this problem, I will continue from where he left and will completely integrate the
Archer> run_simulation command. The main work I am thinking to do will get done in the src/libged folder, the C++ code will get after integrated with the Tlc scripting language, that will call the C++ code for implementing a command. First I would further test the Sphere to Sphere collision and the collision pair generation logic. The btConvexConcaveCollisionAlgorithm, from the Bullet Physics Engine, supports collision between convex shapes and (concave) trianges meshes. This would be a nice addition to the BRL-CAD simulation. Also, in the same folder src/libged are functions that need to be ported to the src/librt level, the
int apply_material (struct ged *gedp, char* comb, char* material, unsigned char r, unsigned char g, unsigned char b);
int apply_color (struct ged *gedp, char* comb, char* material, unsigned char r, unsigned char g, unsigned char b);
that pass the color combination and apply it to a material, showing the current state of the object inside the physics engine. In the same folder there are some more functions that are not properly written and that need some additional work, like the function that gets the exact overlap volume between the 2 aligned axis bounding boxes and some more work on ray tracing.
Arbitrary collision detection and simulation will be the next milestone, in this section I think it would be proper to add some more physics specific attributes to the objects besides velocity that it is already there. I am thinking about gravity, friction force, some bouncing ball effect, like having a ball dropping on a hard surface and bouncing back in the air until it gets in a stable position. Here I will also add ground planes, materials which will be taking account of the surface of the ground. Currently only the bounding box is present in the implementation, but objects will get properties like elasticity and other custom forces. Integrating the Bullet btCylinderShape and completing the implementation of detecting collisions for arbitrary objects would be the next logical milestone.
After the primary physical forces are implemented, I will work on an improved gravitational system, where the gravity would depend mostly on the corps masses, bigger mass bigger gravity. For example if we have large objects dropping on the surface of a plane (e.g. Earth), there we will have to take into account the Earth gravity but also the pulling force that satellites (larger objects that gravitate around the Earth, like the Moon). I am thinking about implementing this as a command, in the archer GUI, after drawing the object the user can use a command similar to
set_objectMasses(Object *obj, type_t Value).
The function will be called and the pulling force, the gravity will be calculated considering the object masses.
So far I have managed to compile and install BRL-CAD on a 64-bit Ubuntu 11.04 version. I am momentarily focusing on the “simulate” command in the archer, looking over the code Abhijit Nandy wrote for it. I have been going through the documentation/resources available on the brlcad.org/wiki. I already have the Sourceforge and brlcad.org accounts. Now the proposed schedule, for the first month of the program, that I have come up with it is as follows :
- Week 1st - 6th August - Week 6th – 13th August
- getting familiar with BRL-CAD simulation, going through the tutorials on this sites : http://brlcad.org/wiki/Main_Page , http://content.gpwiki.org/index.php/BRL-CAD:Tutorials
- getting familiar with the Bullet, going through the Bullet's tutorials and from the Bullet user manual going through the fallowing chapters : Quickstart, Library Overview, Bullet Collision Detection, Collision Filtering, Rigid Body Dynamics, Constraints.
- getting familiar with Tlc scripting language
- start testing the sphere2sphere collision
- Week 13th – 20th August – Week 20th – 27th August
- continuing testing the spehere2sphere collision with convex shapes
- adding the functionality of the btConvexConcaveCollisionAlgorithm, from the Bullet Physics Engine, to BRL-CAD, implementing concave-convex collision
- Week 27th August – 3rd September
- testing the new concave-convex collision with as many objects as possible
After this is done, further discussion with the mentoring organization will set the course of action that I must take.
Here I will log the changes to the project as I will make them.
Starting reading on the BRL-CAD documentation and going through the tutorials from this page http://brlcad.org/wiki/Main_Page.
Going to the tutorials from http://content.gpwiki.org/index.php/BRL-CAD:Tutorials from the following sections: Geometry Modeling Kernel, Geometry Conversion, Procedural Geometry.
Continuing going through the tutorials from http://content.gpwiki.org/index.php/BRL-CAD:Tutorials.
Starting looking through the Bullet library, and going through the "Hello, World!" tutorial.
From the Bullet user manual going through the fallowing chapters : Quickstart, Library Overview, Bullet Collision Detection, Collision Filtering, Rigid Body Dynamics, Constraints.
Continue doing some minor Bullet tutorials, learning about Collision callbacks and Triggers in Bullet.
Starting getting myself familiar with Tcl scripting language, going over this http://www.tcl.tk/man/tcl/tutorial/tcltutorial.html.
Continue the Tcl tutorials.
Starting looking in the BRL-CAD source file, the source files for the simulate command.
Problems understanding some of the simulate command functionalities.
Starting testing on the simulate command with some basic shapes like spheres.
Some problems appeared with the simulate command, not understanding exactly the use of it.
Looked into the simulation system and did some of the simulation presented in the BRL-CAD documentation, also looked into the submitted file in this patch https://sourceforge.net/tracker/?func=detail&aid=3531595&group_id=105292&atid=640804#
Looked into the nirt command and Interactive raytracing- Nirt command. The problem with the simulate command still persist.
Starting working on the improving on the simulate command, adding new parameters (new forces to be taken into consideration). http://paste.ubuntu.com/1165086/
Understanding the way compiling and running a BRL-CAD tool works. First patch with learning purpose submitted, patch # 3562134.
Looking through the rest of the if_*.c files from /src/libfb and modifying all the double-buffer by default options to single buffer. Submitting patch #3562423. Starting looking through the source code for porting the 'joint' command from /src/mged to /src/libged.
Working on the joint command, trying to get it ported by following some of the other get_*() commands, so far did this http://paste.ubuntu.com/1175850/ , but getting some errors http://paste.ubuntu.com/1175838/.
This is what the /libged/joint.c command looks like so far:
- - header file : joints.h http://paste.ubuntu.com/1177572/
- - joint.c : http://paste.ubuntu.com/1177574/
I am having some problems with the debugging of the command, the mged tool exiting altogether when I try to run "joint help" from the /libged/joint.c file. Somehow an "_exit_group" syscall gets called.
31st August - 1st September
At Sean's suggestion I deleted the whole libged/joint.c file and started with a blank ged_joint() function that only printed a message, this was done in order to grasp the way the function calls actually works.
Here are the files :
- header file : paste.ubuntu.com/1178930/
- joint.c file : paste.ubuntu.com/1178931/
The libged/joint.c file has been modified so that the ged_joint() function will print the same output as the f_joint() function from /mged/animedit.c file. When the joint command doesn't receive enough arguments, we are assuming the user needs help with it so a table with all the joint command's arguments it's printed.
Source file joint.c : http://paste.ubuntu.com/1180961/
The libged/joint command is almost done, the files can be found here :
- heder : http://paste.ubuntu.com/1186753/
- joint.c file : http://paste.ubuntu.com/1186752/
- columns.c file in libged library (needed for printing): paste.ubuntu.com/1186754/
The help and debug options for the new joint command are tested and have the same behaviour as the ones from the old joint command (mged/animedit.c). Here are the source files so far:
- joint.h : http://paste.ubuntu.com/1188767/
- joint.c : http://paste.ubuntu.com/1188765/
- and of course the columns.c file http://paste.ubuntu.com/1188769/
Submitted patch #3565411 for the new joint command from src/libged.
Compiled and run the simulate command, from the libged library.
Mged simulate command tutorial : http://brlcad.org/wiki/Mged_simulation
Mged simulate command tutorial is done. Scripts for generating the png images are working.Added explanation about how to create a movie out of the generated png files. http://brlcad.org/wiki/Mged_simulation
Solved a problem with the way the simulation movie was looking, it was way too dark, setting up a default source light did it. Added some *mpgs with the simulation and some gifs of the 2 scenes, one with the viewset the other one without it :
- http://youtu.be/XS1wUnQ8q0M - with the viewset
- http://youtu.be/f-04dvJ7iIQ - without the viewset
- the gifs can be seen on the wiki page : http://brlcad.org/wiki/Mged_simulation
14th - 16th September
Finished the Mged Simulation tutorial.
There was a problem with the "rt" command, the outputted image was way too darker, at Sean's suggestion used the -W option to invert the front and back colors, and the images look properly now.
Modified the second script, now the edges of the geometry are also traced, the view is better and images are rendered using a frame buffer.
19th - 22st September
Started working on improving the simulate command, founded additional problems, not sure yet where they are coming from :
- the simulation for meter units takes way too many steps
- the cube falls through the plane and then comes back to the top(to the original position) and starts falling back to the ground, e.g for the same simulation as the one in Mged_Simulation on the wiki page, with the units set to meters I ran it for 5500 steps first time, the cube was pretty close to the ground, then I wanted to see how closer can I get it, and I ran it for another 800 steps and the cube ended being in this position
- simulating for 300 steps doesn't really give any real movement to the cube, this is wrong it should take less steps for the cube to have a real change in its position
- the velocities are not correctly passed from the bullet, where they are being calculated back to the cube, where they should be updated after every step
- the mass is calculated wrong, it is way too big for a cube, and it gets updated every step, mass should be a constant, for the moment it has been hard-codded to 1kg, that should change
- The cube dimensions are not passed correctly to the bullet and back, after hard-codding those as well there is still a problem, the simulation looks like this with an output for the velocities, cube dimensions like this one http://paste.ubuntu.com/1221285/
- There is also a problem with a variable 'size' shadowing another variable from bullet, so basically I can't add or remove any bu_vls_printf, bu_log lines to the existing code in /src/libged/simulate/simphysics.c
Tried to solve the m/mm problem the simulation had. The simulation acts correctly only for the mm units, if we set the units to m before creating the geometry, then when the geometry dimensions are being transferred between mged and bullet, bullet gets the #meters * 1000 units. I am thinking that mged saves internally the units in mm, even though we are working in meters, so we(Abhi, last year participant on this project and I) came up with scaling the units when they are transferred to bullet with 0.001, but then the mm geometry should be big enough so that the bullet tolerance of 0.04m won't affect the physics. I tested it and for geometry greater than 1000mm it works properly for m and mm.
The expected result is to allow the user to apply different forces to arbitrary objects and be able to simulate as accurate as possible the real world physics. The forces will be applied using an Archer GUI command and the user will be able to visualize the transformations the object is suffering in real time. Adding these facilities to the BRL-CAD program it will give the user the possibility to simulate a real world physics through the software.