SphereCast?

[caxapexac]

Hello, is there any way doing something like sphereCast(radius, startPoint, endPoint, &castCallback)?

I mean you can try to simulate it using capsule overlap but you'll get wrong contact points then

Usecase - moving object wants to cast against wall to move until it and stop. It's not like CCD because we don't need to check time of collision between 2 moving objects, we need cast between static and dynamic object (exactly wide raycast)

[caxapexac]

(the only solution I've found so far is binary search on the way doing overlap several times but it is costly and not so accurate. You can optimize it doing capsule overlap then finding collision point then doing several sphere overlap tests until you find maximum distance to collision)

[DanielChappuis]

Hello,

I think that what you need is called 'sweep test' but this kind of queries are not supported yet in the library.

[caxapexac]

Any plans adding this in near future?

[caxapexac]

Also what can I use instead of sweep test for now?

[DanielChappuis]

Any plans adding this in near future?

Maybe not in the next release v1.0.0 (next 6 months) but probably in the next one v1.1.0. I am not really sure yet.

Also what can I use instead of sweep test for now?

Maybe you can try to use the PhysicsWorld::testOverlap(...) methods that test overlap between shapes where one shape could be your sphere and you test this sphere at several position along a line. That's not perfect of course.

[thomaspeer]

Hi there :) Any update regarding sweeping support as of 2024?

[pauls0416]

It's almost 2025 - any solution to this yet? The only thing I can think of for a shape cast is to fire a BUNCH of rays from the closest half of the primitive all in parallel in one direction by the distance passed in (so, if it's a sphere, for example, the very front of the sphere would touch a flat surface first). This would obviously be very expensive though. You need a lot of rays so you don't miss a thin protrusion and sweep right through it. I really would rather see a better implementation built into the library that utilizes existing architecture, please :) I looked at the source and I don't understand it enough to do something of that magnitude on my own.

[pauls0416]

Daniel, I took your advice and I ended up implementing my own CapsuleCast and SphereCast using PhysicsWorld::testOverlap and advancing the shapes 0.05f each time and it's incredibly slow. I'm trying to make a character controller which needs to sweep a capsule and my solution is prohibitively slow because I'm using your library for server side (simulated) player movement because it's a very authoritative MMO server. It has to handle many players' movements at one time. That's a lot of testOverlap calls. I hope there can be some integrated solution soon that takes advantage of your library to its fullest (I envision something that tests for an entire extruded shape along the whole sweep at once, and only iterates through the colliders once). I tried out Jolt Physics to see if it would be faster but yours wins, hands down! So I'm really hoping I can make this work. While I await a response, I'm going to keep trying to understand your source (i.e. CollisionDetectionSystem.cpp) and try to alter the code but I'm not too hopeful based on what I'm seeing so far.

[haraldwer]

Maybe my implementation will work for you @pauls0416 :)