Radar gives you range rate directly. Getting range rate from vision is hard and takes several frame times at least. If you have position and range rate and dump it into a machine learning system, gesture recognition should be straightforward.
It's not clear how useful this will be for GUIs in general. It could be useful for interacting with large screen displays, where you may be too far back for touch and a mouse is a pain. A presentation version could be useful.
(Presentations from computers still look lame. The audience usually sees the presenter's desktop while they futz with the computer, then enter PowerPoint.)