I'd like to present our project for which the Bullet library is a crucial part. Maybe you can answer some related questions.
We, my colleague Oliver Walter and me, Kai Kostack, are working on an add-on for Blender which enables the user to simulate building collapses in a scientifically accurate way. This is done through creation of physically plausible constraint relationships between rigid body elements and calculation of breaking thresholds based on civil engineering formulas.
This development is part of an EU funded project, named "Inachus". The project in general examines measures to improve the first response in case of catastrophic events like earthquakes or explosions etc. A particular part of this project is the simulation of the effects that catastrophic events have on built structures. We are working for the Laurea University in Finland and we have partners like the Fraunhofer EMI and Applied Science International (ASI) helping us to validate our results. The code is already available and released under the GPL.
Here are two examples:
https://youtu.be/TCJq_2-q34k?list=PLGYn ... 2TSL9w9hTc
https://youtu.be/yu2bQtIS69g?list=PLGYn ... 2TSL9w9hTc
We had the opportunity to present the tool at the Blender conference last year:
Paper: https://inachuslaurea.files.wordpress.c ... xtract.pdf
Add-on for Blender: https://github.com/KaiKostack/bullet-co ... ts-builder
For news updates you can visit our dedicated blog and the official Inachus website:
When testing all kinds of setups you learn a lot about a physics library. While Bullet follows real world physics very accurately and consistent in many ways, there are some phenomenons that we have no explanation for yet.
For instance we have noticed a strong and noisy deviation for angular breaking thresholds of 6dof constraints from the real world. We have tested chains of cantilevers of different dimensions and compared the simulation results to our calculated results. See the attached diagrams for the differences.
Even large changes of the numbers of simulation steps per second or constraints solver iterations didn't reduce the problem. I thought it could be partially caused by precision errors so we tried to compile Blender with DOUBLE_PRECISION enabled for Bullet, which led to changes of the pattern but I would have expected more significant changes (second image).
We would like to know the technical reasons for this behavior for our report.
The time scale of the simulation obviously has influence on the strength of the breaking thresholds indirectly, but apparently it only affects linear forces but not the angular forces in a 6dof constraint. I think this would lead to inconsistent behavior for all time scales != 1. Can this be a bug in Bullet itself?
That's it for now. We'd like to thank Erwin Coumans for this wonderful library.
Kai & Oliver
Show what you made with Bullet Physics SDK: Games, Demos, Integrations with a graphics engine, modeler or any other application
1 post • Page 1 of 1
- Torques with generic 6dof constraint
- test_torque-with-generic.png (137.22 KiB) Viewed 5860 times
- Torques on double precision
- test_torque-with-generic_double-precision.png (135.19 KiB) Viewed 5860 times