Gaussian Fieldwork¶

This is the least defined of the subteams. I have not written any code for it, picked a stack, or decided what the final system should be. I also know less about Gaussian splatting than I do about the other project areas. That gives this team more room to try things and change direction when something turns out to be interesting.

The first work is practical: make some splats and get a feel for the capture process. Try a hallway, an outdoor courtyard, vegetation, reflective surfaces, thin objects, and a scene with people moving through it. Some of those should work better than others. Lets understand why.

Once we have a feel for that workflow, try using the result for something a robot might need. Render a camera view, localize a returning camera, compare depth with a real sensor, add semantic labels, or bring the scene into a simulator. A splat does not need to solve every mapping problem to be useful.