By using floating point numbers you sign contract "I know they are not accurate, but I am ok with it". With
double the innacuracy is so small, that it's practically irrelevant, just to give you an example, you will have 1 cent innacuracy in operations with quadrillions of dollars. But if you going to jump with hyperdrive from one galaxy into another such error in calculated angle may be catastrophical (you may jump into a middle of sun), but then you wouldn't use double.