They're using laser-scanning rental projection, which could hypothetically produce objects indistinguishable from the real world. The biggest two benefits of this are real depth (you can focus on individual objects and it avoids the vergence accommodation conflict) and ultra-high resolution.

I have a source that's told me they've solved FOV, but I can't figure out how their technology could do that from a technical prospective. Hypothetically you don't need a dark room for ML's technology, but their demo video was darkened which makes me wonder if there is some limitation of this.