Page 1 of 1

Different sensor read with different physics engines

Posted: 20 Sep 2021, 09:19
by BlueSkeptical
Hi all, I'm trying to simulate the skin behaviour of a thigh getting hit by a sphere of \(10kg\).
The scene looks like this:
The skin is simulated with a spring/damper joint where K=\(5e^4\).(I found this value in some papers).
The top plate(Plate_dyn) weights \(0.2kg \) while the thigh (UpperLeg_dyn) it's about \(70kg\), in order to simulate the weight of the entire body supported by the leg.
The spring has a length of \(2cm\). Then I plot the force of that Prismatic_joint on a graph.
My question is: Why do I get such differences in the force when I use different physics engine? I know there might be some differences, but here the maximum force read with different engine:
- Bullet 2.78: 880 N
- Bullet 2.83: 365 N
- Newton: -220 N
The differences are huge.

Re: Different sensor read with different physics engines

Posted: 21 Sep 2021, 07:10
by coppelia

it is normal to have some differences. But those differences indicate that something in the modeling is wrong. The first thing that comes to mind is that with most engines, too large differences between masses will result in strange/wobbly behaviour. See the dynamics design considerations 7 & 8 for example.

Another thing that probably plays a role is the thickness of the metal sheet, which appears to be very thin: thin objects can be tricky since a penetration depth (used to compute a reaction force) is difficult to find/interpret.

Make sure that your colliding objects are pure primitives, i.e. cuboids, spheres. If that is not an option, make sure to at least have convex shapes. This directly related to dynamics design considerations 1-3