Where you have i, j , and k, that remain constant along the axes, polar coordinates are just another way of addressing the same problem of denoting a position in space. Suppose, in just using the xy-plane, you have a circle of radius one, centered at the origin. Using Cartesian coordinates, you'll constantly have to change the values of x and y, to satisfy the equation,
√x2 + y2 = 1
But in Polar coordinates, the radius r, remains 1, where it is simply the angle θ that changes as you revolve around. As you said, er is the unit vector in the direction of r, (in this case, r is just 1). And eθ is there TO literally denote the direction of the vector. Rather than breaking down the vector into x, y, and z components, each with its own direction, you simply designate the direction with an angle θ.
I understand your confusion as eθ doesn't have a fixed position, but could give any direction, based on the value. eθ is just there to designate that you're moving in relation to θ, and that it's your coefficient in front that denotes the magnitude of that direction.
For example, if you have a point at (1i,1j), you have a value of 1 along the x-axis, and a value of 1 along the y-axis. Converting this to polar coordinates, you have r=1 and an angle of (pi/4) which turns to (1(er),pi/4(eθ)).
eθ is just there, denoting that you're changing the angle with respect to the positive x-axis.
I hope this makes better sense. If not, please ask any questions, or address what you're having trouble with so i may attempt to help.