USING ADJUSTABLE VISION COMPONENT FOR ON-DEMAND VISION DATA CAPTURE OF AREAS ALONG A PREDICTED TRAJECTORY OF A ROBOT

Implementations set forth herein relate to a robot that employs a stereo camera and LIDAR for generating point cloud data while the robot is traversing an area. The point cloud data can characterize spaces within the area as occupied, unoccupied, or uncategorized. For instance, an uncategorized space can refer to a point in three-dimensional (3D) space where occupancy of the space is unknown and/or where no observation has been made by the robot-such as in circumstances where a blind spot is located at or near a base of the robot. In order to efficiently traverse certain areas, the robot can estimate resource costs of either sweeping the stereo camera indiscriminately between spaces and/or specifically focusing the stereo camera on uncategorized space(s) during the route. Based on such resource cost estimations, the robot can adaptively maneuver the stereo camera during routes while also minimizing resource consumption by the robot.