[Nav2] 3D obstacle avoidance with Realsense D435(i)

edit

Is it really the only answer to use the non-persistent voxel layer or STVL for Realsense 3D pointclouds? And if so, why is almost every example in the internet using the normal voxel layer?

It's an answer, there are several, including building your own 3D perception layer. The examples you find online are likely much older than STVL. I've made some documentation / tutorials on STVL but that was only released in 2018 while the Nav stack has been in business for about a decade prior to that. So you'll find many more resources from that time and from momentum as the "default" in the Nav stack. It's also not a general purpose replacement, it really has a niche it is built to support requiring pretty decent sensor coverage and/or operating in reasonably dynamic environments.

Don't fear, there are many companies using STVL and its stable (since I rarely hear complaints) so its a decent option to consider -- especially if you have good sensor frustum coverage like your describe.