This question was raised on Facebook a few hours ago. Some say it's possible, others say impossible. What do you experts say?
AROS is an existing reimplementation of AmigaOS 3.1 with cross-platform compatibility (current distributions run on MC68020, PPC and x86 platforms; given the diversity of these platforms I would presume adding ARM compatibility would be fairly simple). While its developers have stated they have no desire to implement the updates that were included in the more recent AmigaOS 4 release, it would serve as a useful base for a fork.
Wikipedia has a list of the most major changes in AmigaOS 4. There's nothing here that's Earth-shatteringly difficult. As OS development tasks go, it's not trivially easy but it isn't really hard either. Of course, working on an OS is never quite as simple as working on standard applications, as you have to either deal with emulated hardware or running on a system that may not have a functioning OS while you test it. I reckon a small team with people with experience in the relevant fields (low level hardware interfacing, on-disk data structures, UI and graphics development) could do it in about a year. So from a technical perspective, I'd say it's clearly achievable.
But then, the AmigaOne platform and consequently AmigaOS 4 were commercial failures, so the need for such a system doesn't seem to be there otherwise it wouldn't have failed. So finding a team motivated to spend the time to make these changes would be tricky. As would getting funding to pay such a team. So from a practical perspective, I don't imagine it will ever happen.
In addition to Jules's AROS reference: given the functionality of both x86 and ARM platforms and AmigaOS 4's hardware requirements, it's definitely possible.
However, for a native platform port you not only need to make the bare OS run on the hardware, you also need to provide drivers for the pool of hardware you need to work. With AmigaOS's hands-on approach, writing the code isn't very hard but nowadays, getting to the hardware details is - there's only few hardware around where sufficient hardware documentation is provided.
All in all, the difficulty isn't very high if you're able to stick to hardware that's well documented. Nonetheless, you're looking at several developer years of work.