I imagine building out supply lines for refueling, developing sophisticated position tracking methodologies, and lots of other neat problem solving with no risk to human life.
Even if they could be fully autonomous, there's still a bunch of issues, since caves in general and submerged ones in particular are GPS-denied environments. Acoustics like USBL equipment probably wouldn't work in a cave, instances of underwater VSLAM are still pretty nascent, and inertial navigation system with a low-enough error accumulation rate are almost universally export-controlled by their respective countries.
It's definitely an interesting problem, though, and I honestly think that progress in the area is going to be made because folks that aren't experts in caving end up getting interested in it. For example, I'd imagine this kind of a problem is something that space agencies are also really interested in, since there's a lot room in the middle of their respective Venn diagrams.
Stone's DepthX AUV is a fully autonomous unit that was intended as a proof of concept for mapping Jupiter's moon Europa. https://stoneaerospace.com/depthx/
It seems a lot could be accomplished with a tethered probe. A tether solves a couple of problems, namely power and data retrival. Granted, a tethered probe is unlikely to climb or fly.