Version Roadmap

[Norbo]

I do have MSE (which, along with MBAM, did indeed catch it after the fact), but unfortunately I can no longer trust the harddrive due to the type of work I do on it and the risk of reinfection/incomplete removal.

But yes, it's always nice to have a fresh install

[Norbo]

A quick progress update on v0.15.0:
The "fixing countless bugs and adding final features" phase has begun. Once the engine can run as a whole, I'll start posting preview versions of it here on the forums. It'll still be very buggy and missing a few things initially, but it will help with the transition and could catch a few bugs that would otherwise make it to release.

[Danthekilla]

Sounds great
We will report anything we find with it.

[Norbo]

Like always, it's taking longer than I'd like to get things to a workable point. Progress is continuing though, as always

The 'very early' collision detection tests show promise though. I don't have any direct benchmark numbers, but the general convex collision case seems to be a lot closer to the special cases in speed (as expected).

[Danthekilla]

Sounds good

Great to here what's going on.

[Norbo]

I'm looking to have an initial version posted up here by the end of the night. It's going to be missing a lot of features, but the core stuff (general and special cases for convex shapes, collision response, etc.) is there and working.

Vehicles, characters, static meshes, terrains, instance triangle meshes and other extensions will be some of the notable exclusions. Some of them are implemented and just need a couple of bugfixes (vehicle, simple character controller), and others need more work (non-simple character controller, non-convex collision shapes). Finishing the non-convex collision handling is the next thing on the to-do list and should go pretty fast since all of the 'hard' part has been worked out during the convex implementation.

I have done a little more performance investigation, and it does indeed appear that the general case collision detection system is faster than v0.14.3's box-box special case. On the PC, a 25 ms simulation using the box-box special case runs at around 20-21 ms using the new general case. That jump in performance in the general case should be particularly noticeable on the Xbox360/WP7. The box-box still has better numerical behavior, and I think I can speed it up a bit too.

And I apologize for the repeated delays

[Danthekilla]

AWESOME!!!

But seriously that sounds great.

So the box vs box is around 25% faster, that is a nice improvement considering that the box vs box was already very fast.

Is that because of broadphase improvements or improvements to the actual box narrowphase?

What kind of improvements can we expect on the general case like capsule vs box or sphere?

Static tri meshes are missing at the moment but when they are integrated what kind of improvements can we expect with things colliding with them.

[Norbo]

I'm not sure how much faster the box-box will be compared to the old box-box quite yet (it may be 5-10% faster when I get through with it); the numbers I posted were for old box-box (25ms) to new general-case (20-21 ms). I don't have a comparison directly between old general case and new general case, but you've probably noticed how much slower the general case was than box-box in previous versions, especially on xbox/wp7. Now, the general case has leapfrogged the box-box (at least temporarily), so it should be pretty noticeable

I'll probably still use the box-box special case for boxes due to its robustness since its performance shouldn't be too far off from the general case's by the end. Of course, it can always be turned off too.

That particular boost was almost entirely due to narrow phase changes. The convex-triangle case will be substantially faster than the convex-convex or box-box cases in most common cases. In the worst case, it will be equivalent to the convex-convex case because it actually falls back to the convex-convex case when the triangle and convex are in certain configurations.

The mesh collisions which will use the triangle-convex case extensively also have a few extra tricks to save solving time, so it should be faster in that respect too.
All of these changes are also associated with improved behavior, as well.

The static triangle mesh performance improvements will be pretty significant, since they rely almost exclusively on the new triangle-convex special case. In addition, the architecture has been redone so that the meshes multithread better (the whole 'first class broadphase citizen' stuff). This is really too early for me to make a really good guess, but I'd put the potential speedup for mesh collisions in the 2-8x range. Once I get something up on the Xbox360 or WP7, I'll be able to narrow that range.

It might be a little hard to do comparisons directly between v0.14.3 and v0.15.0 with the upcoming initial release due to some lingering diagnostics code or disabled codepaths which would affect performance, but those will be resolved as it nears completion.

[Danthekilla]

Ahh sorry I misread your post.

But the general case running faster than box vs box... thats incredible

Are there any enhancements being made to the broadphase in this or future versions? Does bepu use a quadtree for it or a 3 axis sweep and prune or what?

At the moment Sphere to box and Sphere to Sphere don't exist do they? Are there any plans to encorporate more special cases or is the general case just that good now that it would be of little benifit?

Thanks for letting me know whats going on, I love hearing about all this.

[Norbo]

Are there any enhancements being made to the broadphase in this or future versions? Does bepu use a quadtree for it or a 3 axis sweep and prune or what?

The primary broadphase is a dynamic hierarchy that reconstructs itself on the fly as stuff moves around. v0.15.0 leaves the broadphase mostly unchanged, apart from its queries and output format. There are plans for both fairly significant improvements to the dynamic hierarchy, as well as some re-exploration of some other types of broadphases. This is scheduled for v0.16.0.

At the moment Sphere to box and Sphere to Sphere don't exist do they? Are there any plans to encorporate more special cases or is the general case just that good now that it would be of little benifit?

Sphere-box and sphere-sphere both exist and both are quite a bit faster than the general case. They were also improved a little bit in v0.15.0, though the difference will be minimal. More special cases are planned, like the triangle-convex one and some capsule-related ones, and perhaps a few other simple/common pairs. v0.15.0 will include the triangle-convex, and the rest will come in v0.16.0 and after.

The super-early version is now available over here too:
http://www.bepu-games.com/forums/viewto ... f=5&t=1050

[Danthekilla]

Cool thanks for all that, Ill have a play with the beta/alpha now.
Keep me informed.

[Norbo]

Updated the roadmap a little and uploaded a new upgrade info pdf.

[chronicacid]

Sweet - Can't wait for see what the More Robust Character Controller looks like. Exciting stuff.

[Norbo]

That reminded to update the roadmap. There's a bit more detail now, and the schedule is a bit longer to accommodate the larger number of features. Fortunately, since I'll be pushing stuff to a development fork on bepuphysics.codeplex.com, you won't have to wait that long to try out new features as they're added

[Norbo]

The fork is now available and includes a few bugfixes and the triangle-convex improvements.