Collimating a Newtonian telescope basically means aligning the optical axis of the primary mirror with the optical axis of the eyepiece. An easy way to do this is to use a collimator, e.g., a laser collimator. The collimator is basically a tube that contains a laser that you put in place of the eyepiece. You adjust the angle of the secondary mirror in the telescope so the beam hits the center of the primary mirror, and then you adjust the angle of the primary mirror so that the beam comes back on itself. But it is crucial for this procedure with a laser collimator that the collimator be itself collimated, i.e., that the laser's axis be aligned with the tubing that the laser is in.
Now, if we were writing a philosophy paper, at this point it would be very tempting to say: "And a vicious infinite regress ensues." But that would too quick. For a laser-collimator collimator is very simple: a block of wood with two pairs of nails, where each pair makes an approximate vee shape. You then lay the laser collimator on the two vees, aim it at a fairly distant wall, and spin it. Then you adjust the adjustments screws on the laser collimator until the beam doesn't move as you spin the collimator on its axis, at which point the laser is collimated to its housing. Moreover, because of how the geometry works, the vees don't need to be very exactly parallel--all the work is done by spinning. So, it seems, the regress is arrested: the laser-collimator collimator does not itself need collimation.
Potential lesson: Perhaps sometimes we philosophers are too quick after one or two steps in a regress to say that the regress is vicious and infinite. For sometimes after two steps, the regress may be stopped with a bit of cleverness.
Well, actually, that's not quite right. For the double-vee collimator depends on the laser collimator's housing being a cylinder. And one might argue that manufacturing an exact cylinder requires a procedure like collimation. For suppose that we manufacture the cylinder by taking a block of aluminum, spinning it in a lathe and applying a lathe tool. But to get an exact cylinder, the lathe tool needs to remain, at the end, at an equal distance to the lathe's rotational axis. So that's another alignment procedure that's needed. I don't know how that's done, being foggy on the subject of lathes, but I bet it involves aligning some sort of a guide parallel to the lathe's rotational axis or by moving the workpiece parallel to the rotational axis. So another collimation step will then be needed when manufacturing the lathe.
And so the regress does continue. But still only finitely. At some point, parallelism can be achieved, within desired tolerance, by comparing distances, e.g., with calipers. There is still a collimation issue for the calipers, but while previous collimations involved the spatial dimensions, the collimation of calipers uses spatial and temporal dimensions: in other words, the calipers must keep their geometrical properties over time. For instance, if one sets the calipers to one distance, and then compares another, the caliper spacing had better not change over the amount of time it takes the calipers to move from one place to another. So calipers allow one to transfer uniformity over time into uniformity over space.
But how do we ensure uniformity over time? By using a rigid material, like hardened steel. And how do we ensure the rigidity of a material? This line of questioning pretty quickly leads to something that we don't ensure: laws of nature, uniform over space and time, that make the existence of fairly rigid materials possible. And if we then ask about the source of these laws and their uniformity, the only plausible answer is God. So, we may add to God's list of attributes: ultimate collimator.
There are, of course, other ways of manufacturing cylinders than by using a lathe. One might cast a cylinder in a cylindrical mould--but that just adds an extra step in the regress, since the mould has to be manufactured. Or one might extrude a cylinder by pushing or pulling the material for it through a circular die. In the latter case, one still has to make a circular die, perhaps with a spinning cutter at right angles to a flat piece, and one has to ensure that the material is moved at right angles to the die. So one has changed the problem of parallelism into the very similar problem of aligning at right angles. And I suspect we eventually get back to something like rigid materials anyway.
So the lesson that sometimes regresses stop after one or two steps is not aptly illustrated with the case of the collimator. That regress is still, perhaps, finite--but it goes further back, and eventually to God.