Add first person view model example

[janhohenheim]

Objective

A very common way to organize a first-person view is to split it into two kinds of models:

• The view model is the model that represents the player's body.
• The world model is everything else.

The reason for this distinction is that these two models should be rendered with different FOVs.
The view model is typically designed and animated with a very specific FOV in mind, so it is
generally fixed and cannot be changed by a player. The world model, on the other hand, should
be able to change its FOV to accommodate the player's preferences for the following reasons:

• Accessibility: How prone is the player to motion sickness? A wider FOV can help.
• Tactical preference: Does the player want to see more of the battlefield?
  Or have a more zoomed-in view for precision aiming?
• Physical considerations: How well does the in-game FOV match the player's real-world FOV?
  Are they sitting in front of a monitor or playing on a TV in the living room? How big is the screen?

Solution

I've added an example implementing the described setup as follows.

The Player is an entity holding two cameras, one for each model. The view model camera has a fixed
FOV of 70 degrees, while the world model camera has a variable FOV that can be changed by the player.

I use different RenderLayers to select what to render.

• The world model camera has no explicit RenderLayers component, so it uses the layer 0.
  All static objects in the scene are also on layer 0 for the same reason.
• The view model camera has a RenderLayers component with layer 1, so it only renders objects
  explicitly assigned to layer 1. The arm of the player is one such object.
  The order of the view model camera is additionally bumped to 1 to ensure it renders on top of the world model.
• The light source in the scene must illuminate both the view model and the world model, so it is
  assigned to both layers 0 and 1.

To better see the effect, the player can move the camera by dragging their mouse and change the world model's FOV with the arrow keys. The arrow up key maps to "decrease FOV" and the arrow down key maps to "increase FOV". This sounds backwards on paper, but is more intuitive when actually changing the FOV in-game since a decrease in FOV looks like a zoom-in.
I intentionally do not allow changing the view model's FOV even though it would be illustrative because that would be an anti-pattern and bloat the code a bit.

The example is called first_person_view_model and not just first_person because I want to highlight that this is not a simple flycam, but actually renders the player.

Testing

Default FOV:

Decreased FOV:

Increased FOV:

Note that the white bar on the right represents the player's arm, which is more obvious in-game because you can move the camera around.
The box on top is there to make sure that the view model is receiving shadows.

I tested only on macOS.

---

Changelog

I don't think new examples go in here, do they?

Caveat

The solution used here was implemented with help by @robtfm on Discord:

shadow maps are specific to lights, not to layers
if you want shadows from some meshes that are not visible, you could have light on layer 1+2, meshes on layer 2, camera on layer 1 (for example)
but this might change in future, it's not exactly an intended feature

In other words, the example code as-is is not guaranteed to work in the future. I want to bring this up because the use-case presented here is extremely common in first-person games and important for accessibility.
It would be good to have a blessed and easy way of how to achieve it.

I'm also not happy about how I get the perspective variable in change_fov. Very open to suggestions :)

Related issues

• Addresses parts of Add examples demonstrating how to follow the player with a camera #12658
• Addresses parts of Add tests, examples and benchmarks for RenderLayers and TargetCamera rework #12588

[robtfm]

i think this will cause weird shadow behaviour where the shadow cast by the arm onto the world won't have the same size as the arm itself unless the FOVs match?

i would imagine the more normal approach for this would be to disable arm/"view model" shadows entirely, which would avoid relying on the light/layer behaviour. alternatively you could scale the view model somehow to make it an appropriate size for the FOV and then put it in the base layer directly.

[janhohenheim]

@robtfm maybe I'm misunderstanding you, but your suggestion is already the case. The view model does not cast shadows. In a real game, you would additionally have an invisible model representing the whole player body that casts shadows (not included in this example).

My comment about the lights referred to me wanting to ensure that the view model is receiving shadows.

I also have an alternative implementation that uses a vertex shader to change the view model's FOV. There are multiple issues with this other approach:

• It is prone to culling issues
• It requires copy-pasting Bevy shader boilerplate
• It looks like magic to the uninitiated, which I'd like to avoid for a common use-case

From what I researched, the dual-camera approach used in this example is what is preferred in other game engines as well. If anyone reading this has experience implementing first person games in other engines, feel free to correct me :)

[killercup]

Looks good! Left some small suggestions

[janhohenheim]

Thanks @killercup. I've incorporated all of your suggestions :)

[killercup]

LGTM with CI fixed

[alice-i-cecile]

Awesome! Very pleased to see more pragmatic, task-focused examples.