Actually, I have a trick for exactly this. Start playing in a domain as closely related to the work as possible (i.e. explore a tangent you're curious about),

Actually, I have a trick for exactly this. Start playing in a domain as closely related to the work as possible (i.e. explore a tangent you're curious about), then, if/when you gained some momentum try to identify a good stopping point try to smoothly pivot into what you should really be doing instead of stopping. It doesn't always work, but every now and then it does.