Graph-based VSLAM needs loop closures to minimize the long-term drift. For most studies, the SLAM datasets are quite complex, and have many loop closures.
However, if we try to SLAM a longe, straight indoor corridor (with texture). If we let the robot move from the start to the end of the corridor, then move it backwards to the start, does the revisited positions (loop closure) optimize the result? I do not see many studies let the robot move backwards, so curious about the effects.
Is there any change if we move in the same or the reverse direction? Assume the camera is mounted at the nose of the robot.