Sorry for the delayed reply, as I said above, I am suffering from some debilitating health issues right now. It takes me a while to have the energy to get back to work. I have taken your advice and put together some parallax backgrounds and they are working nicely, thank you. I am using the final version of GM:S 1.4 by the way, sorry for not mentioning that, it slipped my mind. All of your advice is sound and I thank you for it, though I have done most of what you suggested before creating this topic(excluding only the chunking system and the hub world concept). I should have also mentioned that, forgive me please. It seems I need to get better at detailing exactly what I want. I was trying to avoid an overwhelmingly long initial post that would put everyone off. Anyway, let me detail what I do know for you so that I can provide better context for advice.
My aim is for very large rooms that exceed beyond the limits of the room sizes in GM:S. My primary reasons for this are the high quality/large scale of graphics I want to implement, and the size of traversal space I would like to have with it. I am an animator by trade, so I like my HQ graphics.
I have implemented state machines, and I am prototyping several different kinds of AI with it. One of these AI is an advanced combat state machine that makes decisions based on the context of the environment. It can also avoid collision with all objects which brings me to my reason for questioning which of the movement types is best. I would like my AI to avoid collisions as intelligently as possible and dodge incoming fire. Your move_towards_point system suggestion is very sound advice for simplicity, but I am aiming for more complexity. Can I make a system that avoids collisions and dodges incoming fire with the move_towards_point system you described? I have tried and looked into it but have had no good results. I could very easily be missing something. I would appreciate any direction with this you could provide as well.
I have used your method for pathfinding with mp_grids and mp_potential_step for a long time now. It is one of the first tutorials I followed when I picked up GM:S, and I appreciate it very much, by the way. So far it works the best for most of the top down prototypes I have put together. I have not tried to use this with the offset system because I cannot imagine how that would work. Perhaps I could move it with a more simplisitc system until it enters the grid? It seems like it will only work with world chunking if I do not generate anything in negative grid space. Some advice on this would also be very much appreciated.
I have also tried steering behaviors. I have tended not to use them because I cannot get them to reliably avoid obstacles and I have not solved how to prevent the jittering it seems to cause. I am still looking around the forums for advice on this. It seems as though this would be just fine for a world chunking system if I could solve said issues above, but I am not sure how this could work with the offset solution. Maybe I could do what I suggested with mp_grids and mp_potential step and use a more simplistic movement system until I was close enough? I am very unsure about this one.
Finally I have been trying to find any kind of code example of context based steering behaviors I can because of their better ability to choose a movement direction than normal steering behaviors, but I cannot find one. I have read a lot about them and have even read the "Game AI Pro" chapter on it, but I am still struggling to come up with how to code it. Even if I did figure it out I believe I would still run into the same questions I am having about steering behaviors and mp_grids.
That is the exhausting list of why I started this thread. Sorry for it all being so long.
So you have answered my question about what methods I should use to create the worlds (either world chunking or an offset method), but I still have quite a few questions as you can see. Thanks to anyone who reads this and bigger thanks to anyone nice enough to respond or offer help.