Solver stability in an integrated framework

Post Reply
Dr. DOX
Posts: 14
Joined: Sun Jul 10, 2016 11:51 am

Solver stability in an integrated framework

Post by Dr. DOX »

Hi

I'm working about bullet integration for a 3D OSG engine based. At application level, I'm working about a forklift simulation example. This is implemented like the Hinge2Vehicle example, using cylindric shape for wheels.

My framework has variable DT and I'm tipically use the BT_SEQUENTIAL_IMPULSE_SOLVER stepping the simulation with 100 SUBSTEPS and 1./360 TIMESTEP in order to make the constraint solve stable. For example, if I'm use the MLCP_SOLVER (DantzingSolver) with SAME configuration in the Hinge2Vehicle example, my forklift bounce a lot and during the march sometime somersault (without hitting any object obviously). So, I'm using the previous parameters.

Looking to the forklift operation outside the vehicle, the behavior seems good. Attaching a camera to driving position we can look some "dantzing" (please, look at video attached). When a load is taken, there is dantzing again.

I would to ask if you can clarify me the solver topic, in particular:

1. I'm not able to figure out why the MLCP_SOLVER (DantzingSolver) is a lot unstable in my system (stable a lot in the Hinge2Vehicle example): maybe my DT is too high? It would works better with a fixed one?

2. Even with more aggressive SUBSTEPS and TIMSTEP, the problem don't goes away.

VIDEO (please, download videos for higher quality, so you can appreciate the dantzing better):

https://drive.google.com/open?id=0Bwklg ... Wp0Y1Bjams

https://drive.google.com/open?id=0Bwklg ... U5IMUktQ0E

Moreover, the solver stability is the main topic in the vehicle stability also, I think. In order to make the forklift stable on its wheels, I have to make wider wheels. Please, see follow video.

Wider wheels (stable, with previous problems...):

https://drive.google.com/open?id=0Bwklg ... 0kwNk83WVU

Normal wheels (unstable):

https://drive.google.com/open?id=0Bwklg ... GNEZXpNQzQ

There is something terrible wrong in my usage, someone can address to me in a right direction? I've tried axisSweep3 and Dbvt broadphase algo, but the behavior is the same...

Many thanks, Bye

Dario
Dr. DOX
Posts: 14
Joined: Sun Jul 10, 2016 11:51 am

Re: Solver stability in an integrated framework

Post by Dr. DOX »

Added others explanations....some ideas? Many thanks! Bye!
Post Reply