Another technique not mentioned yet: compss augmented with local magnetic field features.
Inertial navigation and maps.
See the paper, but TLDR: stereo video depth sensing combined with fading previous measurements (so if you detected an obstacle at 1 o'clock, 2 meters out 1 second ago, it is now very likely at 2 o'clock, 1.3 meters out. This way you can limit yourself to just detecting things 2 meters out)