A pair of selsyn motors looks like two synchronous motors connected to each other with three wires. They are both connected to AC power too. They remain idle and don’t turn until you turn the shaft of one whereupon both shafts turn. The torques on both shafts are the same. So is the angular position of the two shafts. See diagrams here. Paragraph 10B1 in the above is a concise description of the theory. This appears to be practical information about selsyns.
Selsyns are a curious early 20th century contraption somewhere between a power device and an information device. They are not expensive and one minute to play with a pair where you can turn the two shafts by hand confirms that the selsyns provide a virtual solid but somewhat rubbery mechanical shaft, which is really composed of electric wires. In most engineering design an electrical circuit is for either power or information. The wires that connect the selsyn rotors are a curious combination of both. While the literature distinguishes selsyn ‘transmitters’ from ‘indicators’ the selsyn theory and wiring diagrams do not. Indeed like a real shaft, the selsyn pair can move information or energy in either direction, or in both directions at once. This raises questions when you decide to replace the wires with other information technology, such as digital, or even software transmission. Indeed how can a single selsyn motor connect a real shaft to a virtual one? I think that selsyn theory is strategic because it forces you to think in terms of two way information flow. I think that most robot theory separates sensors from affectors. This is probably often a bad idea. A connection between real and virtual shafts should send information in both directions, just as connections between real shafts.
These issues are familiar to those who design software to do cycle-by-cycle control of motors via power switching transistors. I think the Toyota Prius has such software. Most such applications move information one way, or at most report resistance to motion as an exception, rather than a continuous flow of torque information.
Viewed as a duplex communication channel the selsyn rotor wires can be described as carrying position information in one direction and torque information in the opposite direction. A software implementation may follow this plan. Alternatively with the help of different hardware at each end the signals can be symmetric and represent a linear combination of angular velocity and torque.
Latency in the information channel turns into a wave like effect in the virtual shaft. Neither the real shaft not its digital surrogate gets the signal to the other end instantaneously. I think that classic selsyns cannot amplify torque. Of course gears could be mechanically added to the real shaft but an amplifier would increase the energy flowing in one direction and decrease it in the other. This is clearly possible with software versions. Digital waldos do this. Remote manipulators used mechanical linkage to accomplish this but without amplification, I think.
I rode this pedipulator briefly. There were only about 3 degrees of freedom (ankle, knee, hip?). It was 2D only. See report AD0619296 as indexed by the “Defense Technical Information Center”. Perhaps what I saw was a predecessor of this much more intimidating machine.