BEPUphysics

Go get yourself some speedy physics!

There's also a video with funky beats.

The 2 in largestAxis/smallestAxis>2 is eyeballed, I don't know of a good heuristic or math that could give a better number to avoid executing MPR a second time when it isn't needed. One iffy option: for shape pairs with smoothly differentiable surfaces, you can analytically differentiate the pair d...

If I understand correctly, the IKPointOnLineJoint would be what you want- it confines a point on B to a line attached to A (which isn't what the documentation said it did, oops). The IKLinearAxisLimit can be used to control the motion along an axis, effectively creating two parallel bounding planes ...

(as always, a v2 note: with a sufficiently large speculative margin or enabled continuous collision detection, this problem basically disappears.)

Neat project!

There are no rules beyond what is specified in the license (Apache 2.0 for both v1 and v2). Within those bounds, have at it.

I'm not sure what is happening based on the description, but you can check CharacterController.SupportFinder.HasTraction and HasSupport properties.

If you think something incorrect is happening, could you reproduce the behavior in the demos so I can look at it?

Don't have to, glad it worked for you

One thing I forgot to note in that modified version: I scaled the inertia tensor of the sphere links up by a factor of 10. It's not strictly necessary when combined with the other techniques, but it does make things easier to solve.

This is a pathological case for iterative solvers; extremely long high stiffness dependency chains are near impossible to handle without some form of tricks in the main solvers that most game engines use. The good news is that those tricks do exist, and they can work. The main problems are 1) gettin...

However this option seems to be a dead end since I can not Identify a way to strengthen the joint since it has no spring settings. Additionally, based on the classed description I imagine that the joint is not even supposed to be as mobile as it is. SolverGroups like the WeldJoint in v1 are just gr...

and the other with a more modern Radeon R9 380, both using Windows 7

Hm, that's odd- that should work. It's possible I pulled in some win10 dependency in the demos along the way without realizing.

First you'll need to adjust the constraint stiffnesses. They have a default tuning which is fairly stiff for masses of about 1-10 units, so if you are using 200, you'll need to scale the stiffness and damping coefficients: constraint.SpringSettings.Stiffness *= 10; constraint.SpringSettings.Damping ...

A question though: what do you mean by extreme cases? Very big triangles (compared to the colliding objects), or triangles that are big in one dimension but small on other? (i.e. a triangle with base: 1m and altitude 100m) Both can cause issues, though you'll often see more with sliver-like triangl...

I'd just recommend trying some extruded triangles of the desired height and seeing if it works well enough. It's true that v1 triangle tests can have some numerical issues in extreme cases (particularly with extreme ratios), but it often works fine. If it ends up being an issue, you can just subdivi...

(and of course, there may be other issues at play too- by virtue of being so rarely called, v1's deep contact routine has been subject to less inspection.)