I've done that before without the physics behavior here:
It can be done with physics but the torque calculation needs the object's inertia
torque = inertia*angularAcceleration.
But for that you'll also need to manually calculate inertia from the object's mass since the behavior doesn't give it:
inertia = mass * (width^2 + height^2)/12
Also the units of the mass expression of the physics behavior is off.
so to get a correct value you need to divide it by 50 to get a correct value.
I forget if there are any other quirks that make it tricky. :/
I'll try to get a example going over the next few days.