How to handle an unknown/changing number of layers?

[SirLynix]

Hi,

I'm working on a game with a special setting, its world is intended to be very large (too large for floats and I'm not keen to use doubles) and "sparse", it's a space game where people can pilot their spaceships to go to very far from each other, using some kind of faster than light device.

The idea to handle this technically is to have multiple physics worlds, for example the interior of a spaceship would be a physics world of its own (with the interior being pretty much static) and the exterior being sparse "cells" depending on where you are in the universe.

Those cells all have their scene origin (0,0,0) at their center and are created/destroyed at will.

My question is, how should I handle this with Jolt? At first I was thinking of give each cell their own broadphase (so bodies in multiple cells won't collide despite being at the same location), but Jolt has to know how many broadphase layer exists and it's a dynamic count in my case.

Or maybe I should handle this with multiple JPH::PhysicsSystem? What are the costs of this? I can think of an additional cost when an entity has to change layer/cell (a player exiting or entering his spaceship for example).

I know it's a complex subject but maybe you have a tip about this?

[DocAce]

X4 is a space game using Jolt, and it has a similar problem of theoretically infinite travel in each direction.

We've solved this by setting the coordinate center of the physics system to be near the player / camera. If the player moves more than 50km away from it, the system is relocated and all bodies inside it are moved to match. Objects beyond ~100km don't have physics bodies and are generally simulated in a state we call "low attention".

Very early on we tried to have a grid of physics systems, but that completely broke down at the boundary between two of them. It was not possible to have an object at the edge of one collide with another at the edge of the other, unless you add duplicate bodies that are close to the edge to both of them, which is just inviting lots of unnecessary trouble.

[SirLynix]

Thanks for the reply! However I forgot to say I'm working on a multiplayer game and my issue is mainly on the server, where all active physics cell co-exist. So technically I have an undefined number of "physics dimensions" (cells) at the same position, but none of them should affect others.

[jrouwe]

DocAce has a lot more experience than I have in making space games, but I just want to comment on this:

At first I was thinking of give each cell their own broadphase (so bodies in multiple cells won't collide despite being at the same location), but Jolt has to know how many broadphase layer exists and it's a dynamic count in my case.

There is only 1 broadphase per PhysicsSystem so you can't do this, I'm assuming you meant 1 broadphase layer per cell. I guess this could work, but the number of layers is fixed at initialization time and that is not easily changed. I think the better solution is to use 1 PhysicsSystem per cell. Note that this would involve destroying and recreating the body when it crosses the cell boundary (bodies are tied to their PhysicsSystem), but besides that the amount of work is comparable to moving a body from one layer to the next.

[SirLynix]

Thanks for the reply.

Indeed I was talking about broadphase layers, my bad.

I see the upper limit of broadphase layers seems to be 254 and an empty broadphase layer doesn't seem to cost much (compared to a PhysicsSystem).
I suppose having rigid bodies asking every collision step if they should collide with one of the 254 layers is not really interesting, and having multiple systems allows me to tune their settings depending on game conditions.

I think I'll try this. Thanks a lot!