Throwing my idea out there:
What if the to propellers are synched.
Depending on desired direction, they are "un-synched" ever so slightly.
This causes additional thrust on one side of the "duct".
I wonder if this maybe has two contra rotating single blade props that they can control quickly enough to be able to speed them up for half a revolution to generate asymmetric thrust to provide pitch and roll?
But it could still be based on the same idea: if the blades are not flexing in the same way under acceleration, any asymmetry might do it, if it can be properly controlled.
Which brings us to how I thought this thing does pitch/roll until I noticed that I just got hung up on a difference between a prototype in the picture and a later iteration in the video: contrarotating propellers will influence each other when blades meet, creating nonuniform efficiency over the duration of a rotation. Three peaks and three troughs, like the legs of miniscule of pressure differential. I don't know where the peaks would be relative to the meeting points, but given the general fickleness of aerodynamics thrust of contrarotating propellers just can't be perfectly even all around the rotation. But that's still three-symmetric. It could become a control option when overlapped it with e.g. an n-symmetric grid where n is not a multiple of three, even if not a good one (mixing it with yaw would be a wobbly mess). The prototype picture shows five arms which is how I got the idea, but in the video it seems to be six/three, so this candidate is out (unless the specifically made a misleading dummy for the video, which I don't think, even though the video is cut in a way that it might be).