[seg9585]

Dead reckoning and double integration from accelerometers produce a large amount of error even during a 2-minute FRC round.
I like the idea of embedded RFID -- maybe they could be hidden in a grid configuration below the carpeting, and a sensor on the robot could update its position as it moves over the RFID.

By setting up multiple repeaters in an atneea you can derive your own geolocation, although at that point you may as well design your own field-relative system. I'm guessing you can get pretty accurate results by placing transmitters on the field boundaries.
You could possibly even accomplish it with a sonar-based transmission system -- derive relative position with respect to certain sound frequencies, all transmitted simultaneously from 4 corners of the field.