Robots that must acquire information about a specific target often face severe occlusion, either from surrounding structures or from the target's own geometry. In such cases, a single camera viewpoint provides only partial information, which is rarely enough for tasks like inspection or manipulation.

The work, described in a new arXiv preprint, proposes a receding-horizon next-best-view planning method. Rather than selecting one best viewpoint, the planner chooses a sequence of views and updates its plan as new observations are made. This allows the robot to gradually build a more complete picture of the target.

The approach is designed specifically for highly occluded scenarios, where traditional single-shot planning may fail. By explicitly handling occlusion, the method could improve robotic perception in cluttered or self-occluding environments, enabling more reliable target identification and interaction.