This thing would need a second motor to adjust the speed change, with an encoder, plus encoders on the input and output ends of the transmission. Something has to control all that.
If you need sudden bursts of high torque, overload the motor while monitoring temperature. You can run electric motors far above their continuous rating for brief periods, and if temperature is monitored, this is safe. That's how Tesla's "launch mode" works.
Not seeing the use case for this thing in robotics.
Or the claim that this would allow for easier human-safety by having the torque, if lowered, not be dependent on logic, but on the actual mechanics, making it possible to stop the entire system if a failure in the torque-change motor is detected?
It's unproven. If you have to haul more mass around, that's worse for energy efficiency.
easier human-safety by having the torque (limited)
There are mechanical 'fuses' for torque limiting. Having a transmission that can downshift into a very high torque mode is probably worse for safety. If torque is directly proportional to motor current, there are lots of standard safety devices for limiting motor current.
> a transmission that can downshift into a very high torque mode
I'm missing something here i guess, how is this different from a motor that can deliver full torque just because the logic chip is driving it at full power? Is it because the motor wouldn't be able to reach such high torque in the first place? (Which might be a point in favor of the transmission.)
If the inception drive is set to low torque a second motor needs to actively set it to a higher torque, which you can guard against by cutting power to the second one while in low torque mode, and having mechanical interlocks that cut power entirely if the shape of the drive changes to a high torque configuration while it's supposed to be low torque. Neither of these require continuous logic.
Let me help: you can, and I have taken part in developing one.
Electric servos for steering wheels use safety critical code, with lot of safety checks, and use direct drive. No need for silly patent for common sense out of the United States.
I'm not a mech engineer, i do code. I'm asking to learn and if you can prove me wrong by teaching something, yay.
So my original point was: it doesn't matter if you use direct drive or a transmission, as both will be controlled by software, and ultimately the safety of that software will determine whether the System is safe overall. The same design principles and safeguards will need to be implemented in both cases to provide the needed integrity.
But I doubt there are any patents on this. I guess it would be illogical to demand vendors to use patents by someone. But there are lots safety regulations on the topic.
Wind turbine gearboxes take a lot of pounding as the wind shifts. Wear is a big problem and not fully understood. The trend is toward gearless systems.
Transmission repair atop a wind turbine is not fun. Power semiconductors are good enough that it's easier to deal with this electrically.
For those saying it has too many moving parts, consider the number of parts found in an 11-speed internally geared hub on a bicycle 
I would be really curious to see if this inception drive could be an alternative to internally geared bike hubs.
Still - I would imagine there are many uses outside of robotics for this transmission.
I suspect if you went to the room of clockwork/gears in Musee des arts et metiers in paris, you could find much of this there, but the point is, you might not find all of it, as one composition.
Thats what I love about mecha: its often like rotational lego, you add it in the right combinations and it does stuff.
(the one which gets me, is the use of diagonally mounted rotators as 'wheels' which can drive sideways: you see that on some forklifts. Its like michael jackson moonwalking for machinery...)
The inner pulley wobbles inside the flexible V-belt and is induced into rotating by precession.
Separation of the inner pulleys controls the effective precession cylinder's inner diameter, the variation of which is accomodated by flexing of the belt: how far it is "squeezed off" and displaced from its outer mount.
Note in the precession animation here how the fast spinning green arrow can be regarded as an input (the wiggle), and the slow spinning of the square-centered blue wheel as an output: there is a many-to-one reduction here, effectively forming a transmission.
Not to be confused with: https://en.wikipedia.org/wiki/Precession
Unfortunately my search has yielded nothing but the same topological(?) traps over and over again to the point where now I just listen to podcasts and entertain my wife with seemingly unprovoked laughter as I chase my grass mulcher around the yard.
The responsiveness probably wouldn't be there for this application, but I'm also curious about whether this could be made to work for a longer range gas/hybrid quadcopter.
However, worth noting that just because a patent is filed doesn't mean it (or all of it) will be allowed. So, wrt your second inquiry -- assuming the tech was filed on -- whether or not it can be designed around will largely depend on 1) what was claimed in the patent (e.g., how broad and thorough the claims are) and 2) what is eventually allowed by the examiner. The latter is largely a product of the prior art in the space.
Still nice drive.
This design is inherently more compact than that arrangement, and the parts are simpler. (I think)