Look up Patrick's comments in
this thread, especially post #38.
Thanks - that makes a lot of sense. Basically, what is being discussed here is a "feed-forward" control scheme, where a model of the motor is used to determine the required modulation index (PWM duty cycle) to achieve the desired current (which then in turn controls the amount of shaft torque). This is often much easier to implement than a feedback system, where to the winding current is measured by the controller (seems simple in concept, but is rather difficult in practice).
Note that the end result is the same - the winding current is limited in order to limit torque.
I mean measuring how far off the motor is from its ideal no-load position the more the difference, the higher torque it is forced on.
The problem with this is that any proper control algorithm (sensored or sensorless) should not allow the rotor to "lag" as more torque is applied.