The projector and the combiner create a virtual image on the other side of the combiner mirror. This image is perceived like a true image.
The eyes need to adapt the focal length to the distance of the virtual image. If the HUD is set to collimate at infinity, the eyes must accommodate at infinity to have a sharp image. If the eyes accommodate on the combiner frame, the virtual image is blurry. Without entering into the details of how the whole eye works (I ignore them) the overall principle is the focal length must be reduced (lens more convex, refracting more) when the object to accommodate on is closer.
Here is a drawing where the projector is not collimating at infinity, to facilitate the demonstration.
At the top, the eye accommodates on the combiner, it adjusts the focal length to form a sharp image on the retina.
In the middle, the eye, still focused on the combiner, forms a sharp image of the virtual image ahead of the retina, the image on the retina (not shown) is blurred.
At the bottom, to have a sharp image of the virtual image on the retina, the eye refocuses, increasing the focal length, now the combiner image (not shown) is blurred as the focal length is not adequate.
If the projector really collimated at infinity, then:
In the middle, the image of the virtual image would be sharp on the focal plane and still blurred on the retina.
At the bottom, the focal length would be increased