[alexhegit]

Summary

• switch depth camera render targets from R32f to RGBA32f
• keep linear scene depth in the R channel and preserve the existing one-float-per-pixel consumer API
• update depth preview comments and camera format tests
• assert both RenderTargetFormat and the actual EPixelFormat

Problem

On UE 5.7's Vulkan backend, URLab depth capture allocates PF_R32_FLOAT and reads it through ReadSurfaceData(..., TArray<FLinearColor>, ...). VulkanRenderTarget.cpp does not implement VK_FORMAT_R32_SFLOAT conversion in that path and asserts:

Unsupported format [100] for conversion to FLinearColor

This crashes the editor when a depth frame is read back on Vulkan.

Fix

Use RTF_RGBA32f / PF_A32B32G32R32F, which UE 5.7 Vulkan supports in the FLinearColor readback path. URLab still extracts only .R, so the ZMQ/SHM and ConsumeFloatPixels() contracts remain one float32 depth value per pixel.

Trade-off: the internal depth render target consumes 4× the GPU memory and copy bandwidth of R32f. The external frame format is unchanged.

Validation

• Regression test observed failing before the implementation change
• URLab.Camera.DepthMode_ConfiguresSceneDepthFloat passes
• Full Linux NullRHI suite: 269/269 passed
• UE 5.7.4 offscreen Vulkan smoke test on NVIDIA A30
• Minimal MJCF imported and instantiated
• RGB render and float32 depth readback produced nonblank geometry
• Depth range: 330.621–401.269 cm, 2,967 valid pixels at 320×240

Environment: UE 5.7.4, clang 20.1.8, NVIDIA driver 580.178.04.