Resolved having 1000+ gameobjects with colliders, over 2 fps (2D)

IM working on a simulation project which is struggling to get off the ground.

The current goal is getting an "Ant" bot that can
-See ahead
-determine where food gameobjects are
-move to that food
-interact with that food
-Instantiate new ants after eating enough food (I know that isnt how ants work)

I have achieved all these steps, but cannot have more than about 15 ants active at a time without frames plummeting. The Profiler tool shows me that alot of time (about 95%) is being spent on physics2D.

As far as I know, physics should only be applying to 'ants' right now, as they are the only object with a rigidbody.

If I had to guess where the problem is, its how my 'ant's see what is ahead of them.
each ant has a child gamobject that is a collider, the ant stores any other colliders that enter and are a 'food', then removes them when they are no longer in sight.

But I have no idea if this is actually the issue, or how I can even approach figuring out why physics is spiking so much. Its not the food (the only other go's in the scene with even a collider) because I removed them and got the same results
Heres what the Profiler is showing me. its all good until some arbitrary number of ants have spawned, and then boom,

 

First of all, the gameobject workflow is not made for that many entities. However, thats not (yet) the problem here, since 15 ants is certainly way too low an amount to cause issues.
(To allow for truly large amounts of entities, in the region of thousands plus, use DOTS. Which is by no means beginner friendly or easy to get into - so i dont recommend that unless you have to.)

I'd like to see your code.. not sure how else i would be able to help you.

Why are you not using raycasts for vision? Thats the most common approach.
Also, reduce the amount of code in FixedUpdate to the absolute utmost bare minimum.

 

a couple things worth trying:
1) replacing the single ant collider with multiple primitive colliders that acheive your shape
2) in your physics settings, untick physics collision detections between the Ant colliders' layer with themselves. also the same for food on food collisions. Of course only if you dont need to detect those collisions.

Also, what is happening before the giant spike in the profiler? instantiating 1 more ant? 100 more ants. 15 ants should not cause an issue as you described them.

 

Since this is on the CPU, If you select a heavy frame and scroll down, there should be a break down below of what API calls are made on a particular frame and what percentage of the run-time they took. Try to find out more info.

 

Im looking into raycasting in order for 'ants' to "see" right now, will try to update how that goes.

the ants currently have 1 polygon collider, I will replace it with a box collider, precision is not really needed here so idk why I didnt do that in the first place.

I think I already have layers not colliding with themselves, thats in the collision matrix correct?

and nothing of note is happening before the large spike, except spawning more ants. to test the breaking point of what im doing right now, I just have each ant spawning a new ant on a coroutine.

 

I dont exactly know how to read the information. and cannot upload the profiler data here, the file size is too large.
if im looking at the correct module, i THINK there is a spike in 'contacts' at the same time as new ants are added to the scene. so maybe the ants are hitting each other as they are created and causing a large spike of physics interactions?

 

would you happen to have a good way to do multiple small angle raycasts? most things covering it just use transform.right or .up but that doesnt help me much, this is more of a topdown view and would need small adjustments and im not sure how to go about that

 

Colliders/raycasts work with minimal objects say <100, for anything higher you should think abstract. You need to compare objects that are close to each other. In that sense if 2 objects pass an axis distance threshold, you only have 2 colliders for each specific comparison A/B, and you could, to save math, just turn on some colliders for close objects where collision is relevant. As for method, not tried it, but interested to try at some point.

 

There is probably many aproaches to this. Since you have your ant as a starting location and a view radius, you could look for solutions for finding equally spaced points on a circle. You'd discard any that are not in the field of view of your ant (which you can do by calculating the angles). Now you'd have equally spaced raycasts within your field of view. You would however need to figure out a way so that you wouldnt have to calculate this every frame - but while thinking about that i realised you could probably just utilize a spherecast / overlapcircle, get all food items in the vicinity, and then calculate the angle to those to determine whether they are in the field of view or not. That should be comparably cheap and easy to implement. I then also found this video where Sebastian Lague did just that, fittingly, for an ant simulation:

(5:35, since the Unity Forum appears to eat up youtube timestamps..)

Also a general piece of advice, which might or might not be obvious: you are never the first to attempt something. So usually, for any given topic you want (at least to some degree) there will be articles, blogposts, tutorials or even scientific papers describing something that will help you, which you will likely find when googling for the broad topic itself, since others who attempted it would have had the same issues - or at least needed a solution that works aswell.

 

I would recommend first diagnosing EXACTLY where your performance issue lies. the time breakdown of function calls in the profiler should tell you. you can click on the time columns to order the rows by time spent if it isnt already.
Then Try simple solutions.
Then try solutions that require calculated backflips.
15 ants with a collider should not demand clever workarounds. Ill have hundreds of 3D colliders and run on mobile without issue.

 

Just want to say im learnign alot in a small amount of time so thankyou guys for being very purposful with your responses, really knocking it out of the park here.

is that true? yeah I absolutely have been modifying transform directly, specifically because I assumed having the rigidbody component would intrinsically be heavier.

 

For reference my current code for having the ants 'see' using raycasts instead of the child colliders

Code (CSharp):

1.     public void see()
2.     {
3.         Vector3 sightEdge = transform.right * senseRange;
4.         for (int Exx = -5; Exx <= 5; Exx++)
5.         {
6.             Vector3 sightPoint = sightEdge + (transform.up * Exx);
7.             RaycastHit2D[] canSee = UnityEngine.Physics2D.RaycastAll(transform.position, transform.position + sightPoint, senseRange);
8.             Debug.DrawLine(transform.position, transform.position +  sightPoint, Color.red);
9.             foreach (RaycastHit2D Ell in canSee)
10.             {
11.                 if (Ell.collider.gameObject.GetComponent<Food>() != null && !plantsInSight.Contains(Ell.collider.gameObject.GetComponent<Food>()))
12.                 {
13.                     plantsInSight.Add(Ell.collider.gameObject.GetComponent<Food>());
14.                 }
15.             }
16.         }
17.         Invoke("see", 1f);
18.     }

which is working okay. Im sort of avoiding the actual issue which, after sleeping on it im convinced is from ants spawning other ants on top of themselves and causing an avalanche of collisions.

I think i understand that if I just make the ants kinematic then they wont cause all those collisions?

 

The immediate issue was in fact that the ants could collide with themselves, and when spawning in caused a huge amount of collisions while trying to separate from each other. Thats the original issue solved, updating the title