Epsilon is your "fuzz factor," since doubles may not be exactly equal. Epsilon lets you describe how close they have to be.
If you were expecting 3.14159 but would take anywhere from 3.14059 to 3.14259 (that is, within 0.001), then you should write something like
double myPi = 22.0d / 7.0d; //Don't use this in real life!
assertEquals(3.14159, myPi, 0.001);
(By the way, 22/7 comes out to 3.1428+, and would fail the assertion. This is a good thing.)