CG films and animations commonly feature highly realistic reflections, which are important for giving a sense of “connectedness” among the objects in the scene. However, the accuracy of these reflections comes with a high cost in processor time and while this is not a problem for films, it severely limits the use of reflective objects in realtime games.
Traditionally, games have used a technique called reflection mapping to simulate reflections from objects while keeping the processing overhead to an acceptable level. This technique assumes that all reflective objects in the scene can “see” (and therefore reflect) the exact same surroundings. This works quite well for the game’s main character (a shiny car, say) if it is in open space but is unconvincing when the character passes into different surroundings; it looks strange if a car drives into a tunnel but the sky is still visibly reflected in its windows.
Unity improves on basic reflection mapping through the use of Reflection Probes, which allow the visual environment to be sampled at strategic points in the scene. You should generally place them at every point where the appearance of a reflective object would change noticeably (eg, tunnels, areas near buildings and places where the ground colour changes). When a reflective object passes near to a probe, the reflection sampled by the probe can be used for the object’s reflection map. Furthermore, when several probes are nearby, Unity can interpolate between them to allow for gradual changes in reflections. Thus, the use of reflection probes can create quite convincing reflections with an acceptable processing overhead.
The visual environment for a point in the scene can be represented by a cubemap. This is conceptually like a box with flat images of the view from six directions (up, down, left, right, forward and backward) painted on its interior surfaces.
For an object to show the reflections, its shader must have access to the images representing the cubemap. Each point of the object’s surface can “see” a small area of cubemap in the direction the surface faces (ie, the direction of the surface normal vector). The shader uses the colour of the cubemap at this point in calculating what colour the object’s surface should be; a mirror material might reflect the colour exactly while a shiny car might fade and tint it somewhat.
As mentioned above, traditional reflection mapping makes use of only a single cubemap to represent the surroundings for the whole scene. The cubemap can be painted by an artist or it can be obtained by taking six “snapshots” from a point in the scene, with one shot for each cube face. Reflection probes improve on this by allowing you to set up many predefined points in the scene where cubemap snapshots can be taken. You can therefore record the surrounding view at any point in the scene where the reflections differ noticeably.
In addition to its view point, a probe also has a zone of effect defined by an invisible box shape in the scene. A reflective object that passes within a probe’s zone has its reflection cubemap supplied temporarily by that probe. As the object moves from one zone to another, the cubemap changes accordingly.