Make two physics objects not collide but do detect collisions in Unity

Question

I have a Unity project in which there is a 2D game world that consists of static colliders to make the geometry solid to the characters that inhabit it. The player is a dynamic collider (with a non-kinematic rigidbody). There's also an enemy character that is also a dynamic collider. Both characters walk over the floor and bump into walls like I'd expect them to.

What I want to achieve is that the player and enemy are not solid to each other, so they can move through each other. I achieved this by putting the enemy and the player on separate layers and setting the collision matrix so that these layers do not collide with each other. The problem I'm having now, however, is that I do want to detect whether or not the enemy and the player ran into each other. I added a trigger collider to the enemy character, it's on the enemy layer which means it doesn't detect collisions with the player.

I thought of making a sub-gameobject for the enemy, put it on the player's layer, add a rigidbody and trigger collider to it and use that to detect collisions between the player and the enemy, but it feels so convoluted that it leaves me wondering if there isn't a more elegant solution for this.

Answer 1

Edit (2020-05-26): Nowadays you can tell the engine to ignore collisions between two given objects. Kudos to Deepscorn for the comment. Will leave the rest of the answer unchanged, and read with that in mind.

Yes, you need to create a child GameObject, with a trigger collider, and put it in a layer that interacts with the player layer.

No, you don't need to add a Rigidbody to the new GameObject, the parent's rigidbody already makes it a dynamic collider, so you will get OnTrigger events.

As a side note, just to keep things organized, if you create a child of the enemy don't put it in the player layer. For example, in the future you might need to disable the player's layer collision with itself. Furthermore, if your player interacts this way with many objects, I'd put a single trigger on the player instead of the enemies, on a separate PlayerTrigger layer, just to keep things simple.

Isn't there a simpler way? Not really. You definitely need non-interaction between the player and enemy colliders, but some kind of interaction between them too. So one of them needs to span two layers, or the whole interaction would be described by a single bool. The physics engine processes lots of information in one go, so you can set all the layers and collisions you want, but during the physics loop you have no further control on what happens. You can't tell the engine to ignore collisions between just two objects. Having only 32 layers, and having them behave in the rigid way they do are actually heavy optimizations. If you are concerned about performance for creating another layer, disable interaction between layers you don't need, like the trigger layer and the floor and walls, or layers that don't even touch.

Your alternative is doing it all by code, which is even less elegant. A single child capsule on the player doesn't sound that bad now, doesn't it?