Deep SLAM

If I try ORB-SLAM and it doesn’t work, I will go ahead and try Deep SLAM.

There’s some really interesting work in this field.

But the basic idea is that instead of using hand-engineered features, like in the case of ORB-SLAM, we have a neural network that learns these features, that SLAM can use to track.

Some implementations

• DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras
  • Code: https://github.com/princeton-vl/DROID-SLAM

Related

• Scene Reconstruction