@Farshid I will try to explain, with my best english (which is not my native language):
There is a fundamental difference between game engines and renderers like VRay, Arnold, and alike. All game engines (lightmaps are not an UE4 exclusivity) must consider the best processing times in realtime for gaming. So, a complex lighting scenario with different types of lights and also with static and movable objects can be challenge to produce a single frame with the necessary speed to be useful for gaming. So, people like to game at 100 frames per second (FPS) means that all lighting calculations are done around 100+ times per second and output to the screen. Static renderers and interactive renderers don’t need that type of throughput time.
So, in order for game engines be that fast they resort to tricks and tweaks in the scene for this to happen. Some of these tricks are even more suitable for some applications in desktop, mobile and VR, because they all have differences in processing power or needs a specific performance constraint (VR needs 90 FPS per eye to be considered a good application - google for the reasons).
So, for gaming, we have different lights to be considered in shape and behavior (static lights, stationary lights and dynamic lights) and also objects which will never move and some that not only moves, but they also deform. In the case of non movable objects, the trick for having shadows calculated fast for static and stationary lights is called lightmaps. The lightmap is a hint for the engine to quickly compute the shadows and the lightmap resolution will control how smooth or sharp that shadow can be. For movable objects, static lights does not produce any effect, but stationary lights will affect them.
What remains is when we have dynamic lights, which in this case the light source changes position over time and can change its intensity aswel, which will produce a completely different shadow. Dynamic lights will need to use only the geometry to create the shadows, which is expensive, so there are lots of optimizations a game engine will need to do in order to extract performance in these situations: only objects affected are the ones near the source, far objects if non-movable will still rely on the lightmap to produce the shadows, since the light source is far, etc.
So, often in games, you will see scenes with a mixed approach on the lighting just for performance reasons, which depending on the processing power might be good in terms of visual, but hardly will be realistic image like a scene that can be made with a static renderer for a static scene. The static renderers like VRay, Arnold, etc you will need to rebuild the frame shot in case you change the camera angle and some renderers have an interactive ability, meaning while you move the camera it keeps computing the frame data until the movement stops and it can compute to the maximum amount of time or number of samples predetermined.
Now, we are getting with realtime raytracing (possible because of new NVidia hardware) the ability to faster do not only shadows, but also reflections fast enough for games and these specific cases will still use a mix of old and new techniques to extract performance. The new interactive GPU light baking will behave according to the settings you will define and use the realtime raytracing hardware to do it faster, and by its nature, you can rely only on the geometry and materials to produce the shadows and reflections, meaning no need for lightmaps. Even with strong hardware, the amount of time used to process the frames will increase, but as it is done on GPU, will be fast enough to allow similar interactive cycles and keeping fidelity. So, this feature are more for archviz then gaming, so you can expect the pipeline to be similar to what VRay and others requires.
So, the conclusion is that lightmap will be required for all techniques where you need to save time and you are responsible to choose where it will be used for that purpose. Otherwise, it will rely in the geometry and materials and expend more time per frame to compute the scene.
PS: with the increased power to be seen in the graphics cards from now and the next years, I can’t see a reason for someone who model and texture objects for use in archviz, to not make the lightmaps properly, because those objects can become a source of income if sold for game projects purpose aswel. Just a thought.