Drone flight controllers can use 3 axis magnetic field sensors combined with their 3 axis gyros and accelerometers, and use various sensor fusion algorithms to

Drone flight controllers can use 3 axis magnetic field sensors combined with their 3 axis gyros and accelerometers, and use various sensor fusion algorithms to produce higher accuracy compass headings which shouldn't have _any_ long term drift, since the magnetometer readings can compensate over time for any short term gyro drift. I'm not sure how accurate the magnetometers are, and how accurate the calculated compass heading can be, but I recall a project where I used a pair 6DOF accelerometer/magnetometer breakout boards to measure the angle change between the body and the grind setting adjustment ring to datalog my coffee grinder, and I was getting 10bit resolution degrees which all _seemed_ to accurately track the small sub 1 degree adjustments when dialing in the grind. Although I'll admit to not taking the time to understand all the math involved to get from the raw 3 axis acceleration and microTesla magnetic measurements to compass heading, I just copy/pasted sample Arduino code which gave me 1 decimal place degrees. But my grinder has 100 "notches" around the 360 degrees of the grind adjustment ring, and my data logger at least looked like it was accurately/repeatably showing changes down in the quarter to half a notch range. So while I don't think $50 worth of stuff from Adafruit will get you <0.01deg/hour gyro drift rate, with magnetic calibrations and sensor fusion you could probably hit at least 0.1deg long term precision/accuracy, so over 10 or more hours you'd have "navigation grade" accuracy. Which is probably "good enough" for anything traveling at less that a couple of hundred MPH. 0.1deg error is about 1km every 600km - so if you were sailing from Sydney to San Francisco (with no GPS or astro nav equipment) which is ~12,000km, you'd be within 20km or so when you got there. Which seems "close enough" for anything except delivering nukes.