Yeah, the basic principle does sound solid. It looks like they’re not even relying on it to neatly work like random filters, but are applying statistical analysis to whatever superposition of speckle patterns comes out of the device.
The level of precision they’re talking about sounds more impressive than I would guess for it, though (1nm over 1um), and I don’t see the connection claimed with optical trapping or ultrafast imaging at all. If it checks out, I expect we’ll hear more in not too long.