My M.S. thesis was essentially an attempt to answer this question. The answer, it turns out, is that the readings you get from single bright celestial sources (

My M.S. thesis was essentially an attempt to answer this question. The answer, it turns out, is that the readings you get from single bright celestial sources (sun, moon) aren't accurate enough to enable navigation primarily from this data, even assuming perfect timekeeping. You can use multiple readings of multiple celestial sources (stars) to navigate reliably, but then you've shifted the problem. With a single reference source your main problem is the sensitivity to error (slight angular errors translate to miles). But if you use starlight your primary difficulty is simply measuring the light sources in an affordable way (i.e. short of taking images in real time and doing real time image processing and graph construction on them).