Why would sensorless not be practical in a vehicle? Generally, vehicle = wheels and wheels = gearbox. With a gearbox, small angular displacements of the wheel correspond to a angular displacements for the the motor so sensorless's startup hunting and low speed resolution are increasingly masked the higher the reduction ratio you go.

@Kartman: what is the "old synchro"?

@linuxfansaysReinstateMonica They go way back. My ancient radar that was commissioned for naval use in 1944 used synchros. Also more often called a Selsyn, back then. (I wonder how many still alive remember that name for them ) Usage of "synchro" really didn't start until sometime during the war and wasn't popular for some time after it.

@linuxfansaysReinstateMonica - i gave enough Google hints. Actually I was thinking of a resolver, but said synchro. A resolver is 3 phase.

@Kartman, for what I know, a resolver gives sin/cos of the angle.

Optical disk coded or the same idea, on rotating wheel?

optical disk encoders are not really suitable for bldc motor commutation

Does anyone "think" really that my answer is "opinion" based? It is really how power brushless motors are controlled!

nxp.com/docs/en/application-note/AN1914.pdf

@Antonio51: the question is, at least partly, opinion based. So answers are likely to be opinion based. My opinion is, for example, that classic hall effect sensors are the most reliable (may be not at top of performance). Aside: I would not have closed this question, every "input method" has pro and cons.

Ok for the fact that OP question is "partly" "opinion" based, But if the "answer" is not, it is also a "good" question. The back-EMF method is not only, the most "reliable", but it is the only that does not need sensor ... because the sensor IS the "motor" ... And if the motor is "out of sevice", it is not because an additional "sensor" is "out", but because the motor is "bad".

So, the last sentence >Sensorless would be great but not practical in this direct drive (no gearbox) vehicle.< ... is out of "sense".

@Antonio51, sorry but I disagree. The paper you linked says that the first step in spinning up the motor is "alignment". What if alignment step fails? I use BLDCs on electric gates where the first movement has to be done in the correct direction (the other direction is impossible because the gate is on the mechanical stop and the motor will not move).

@linuxfansaysReinstateMonica ... Isn't the direction of movement imposed by the user? Moreover, that phase of "alignment" can be as short as one wants ... without "rushing" the "material".

@Antonio51: you can not "impose" the direction, if you don't know where the rotor is.

BLDC motors are "asynchronous" at "starting" ... So, if one "imposes" the sense of the rotating field, the rotor must go in the same direction?

@Antonio51, you don't impose a direction, you excite 1 coil (among 6). The direction the rotor takes, toward that 1 coil, depends on where the rotor is. Moreover, under load, you must give sufficient voltage. But if the coil you excite is not the right one (2 among 6), you get very big currents and unpleasant mechanical and electrical effects. There are methods to determine rotor position without moving it, but those are not described in the paper you linked. And they are not simple.

For moving (asynchronous machine), one must excite (at least) 2 coils ... If not, one doesn't have a "rotating" field. And reducing voltage as "usual". But ok for "huge" loads.

@Antonio51 sorry again, a "rotating field" can only be generated by exciting coils in sequence... in the right sequence. Where I wrote "1 coil" for simplicity, read it as "1 step" (among 6). Yes, each step has always 2 coils together, but that is so for construction reasons and because a magnet has always two poles.