Efficient enemy detection with triggers

Hey there!

currently I am trying to find out the optimal solution for the following problem:

in a 2d tower defense game, when enemies come near a tower (or even moving tank) the tower/slowly moving tank take a defensive stance. when there are no enemies near - tower/tank go back to normal stance.

My initial solution was with an attached trigger to the tower/tank, which execute a code in the OnTriggerEnter function. The problem was, that when the enemies are destroyed in the trigger zone, and because OnTriggerExit was not triggered, the tower/tank still "think" there are still enemies near and remains in the defensive stance. Also there can be a lot (~20+) enemies in the trigger zone, that come in and out from the zone.

What will be the performance impact if I attach a stance script on the enemies and not on the tower/tank? So that when the enemies detect a tower near, an OnTriggerEnter in the enemy scripts is triggered, and they call a function on the tower (which they have a reference to in advance) to inform it that "they are near". That will solve the described problem above, because in the OnDestroy() function the enemies will check if they are in the tower zone and inform the tower that "they die and go out of the zone".. The problem with performance would be, that all the enemies will call the same function (or same two functions) of the tower script: enemyComingIn() and enemyComingOut() all the time. This concurrent calls might slow the game.. And besides, these two functions will call an Animator variable of the tower to be updated.

Is there any better solution? Any opinions and help are very welcome!

 

How about let the towers and enemies be responsible for animating themselves?

Have a class that holds a reference to all the enemies positions. Have the towers request that list every frames, iterate through the positions, calculating their distances from the tower. Break out of the loop as soon as one of them is in range, otherwise it means the tower is clear to go back to normal stance.

Same thing, but the other way around for the enemies.

This should be fairly quick to setup, and I'd just try it and see if it works, before worrying about performance hits.

If anything, you could have half the towers request the list for a given frame, then the other half the next frame, and alternate. They'd call it every 2 frames and it would reduce the calculations to half.

Edit : Alternatively, if all towers/enemies have colliders, you could use this for your detection. Same deal, towers and enemies are responsible for handling themselves. https://docs.unity3d.com/ScriptReference/Physics.OverlapSphere.html

 

Hi, Antoine and thank you for the fast response!

Referencing all the enemies and calculating for every enemy the distance on every frame (or actually fixedUpdate) with Vector2.Distance() would be.. the same as a trigger does, more or less, isn't that right?

 

In update, not fixed update, it's not a built-in physics call.

The same, relative to what? Performance? Detection? For performance, you'll have to try it out. Perhaps someone more knowledgeable will step in. For detection, personally I think it grants better control. You could even use the same system to figure out which target should be prioritized. Plus, like I said, you can manage the number of calls per frame.

 

@AntoineDesbiens I mean relative to both: performance and detection.

Also the detection could be done with raycasting to every enemy instead Distance or magnitude, as well as Physics2d.CircleCast or Physics2d.OverlapCircle.

It suffices to know if there is a single enemy near the tower/moving tank in a given distance or none at all. Ignore targeting and reference to the enemy, I don't need that info - it is just about going to defensive or normal stance of the tower/tank.

So, given these conditions I wonder which approach is the best one:
- check with .Distance (tower, enemy) for every enemy if there is one near the tower and change stance if appropriately.
- check not with Distance but with a Raycast for given max radius to every enemy, if the ray hits one
- check with physics2d.OverlapCircle
- check with physics2d.CircleCast. Is it more performant than OverlapCircle?
- use the "classic" Trigger and OnTriggerEnter + OnTriggerExit

What do you think @AntoineDesbiens ? Can't estimate which is most propriate/fastest approach. Unity profiling is something I don't know how to work with .. yet. So if there is someone that has experience with that "enemy detection efficiency problem", I would love to read his opinion!

 

Maybe @MelvMay as Physics guru might give his opinion here also?

 

Thanks for the answer @MelvMay !

Still have some questions, which you might answer straight away, before I start to learn about the Unity profiler:

• What about OnTriggerEnter2D / OnTriggerExit2D? Is OverlapCollider faster than that? How are actually OnTriggerEnter/Exit implemented? In case of Circle collider ~> CircleCast?
• OverlapCollider will give me a List of overlapped colliders.. that means, if I make the detection every frame in a Coroutine, that will internally call a "new List<Collider2D>()", right? Which might be memory inefficient, right?
  
• Will OverlapCollider consider the Collision Matrix from Physics Settings? So I can separate the layer needed just for this kind of collision?
• Rigidbody2D.OverlapCollider vs Collider2D.OverlapCollider? Does it matter?
  
• What I am really looking for is a function that gives me a boolean - yes/no, there are (not) any colliders overlapping. Don't even need a list. As efficient as possible. Unity doesn't provide such function yet, right?

For info: there will be a single moving tank/tower (the player) and up to 20 simultaneously active enemies in the scene. And it's going to be a mobile game.

Can you help with some answers to the above questions, @MelvMay ? I just don't have the internal knowledge how all the Collider detection is implemented in Unity to pick the right choice, I couldn't sleep last night thinking about that stuff..

 

Here is the script I used last time I did tower defence.

http://blog.boredmormongames.com/2014/07/cycling-through-targets-on-tower.html?m=1

 

Thank you @MelvMay and @Kiwasi !

The script that @Kiwasi is pointing to is very similar to what I initially had. That adding/removing/counting elements is the overhead I wanted to remove.

As I already described, the trigger zone is just for changing stance - defensive when enemies near, normal when no one near. There is another trigger on the tank, that picks and shoots at the targets that come much closer.

@MelvMay : should the OverlapCollider be triggereted in an Coroutine or Update every frame? I guess FixedUpdate as you pointed out is not that good, since the simulation hasn't performed at that point of time.. Also, can I provide a null for ContactFilter2D parameter or is just an Initialized ContactFilter enough, so that the collision matrix is directly applied on the contactFilter, without further customisation of the filter?

Again, thanks a lot for the answers and explanations so far!