|
| |
Vision Robotics Technology
Exploration
The robot creates a map of the environment starting from its initial location by using its stereo vision sensors to determine where there is open space and where there are objects. A 3-D grid of voxels which describes the environment is created by "throwing" rays from the cameras to the objects seen by the stereo cameras. The voxels are 2cm cubes for the robot vacuum cleaner and each contains a value which indicates whether it contains an object, is empty space, or is unknown. The ray throwing causes a voxel to be empty if it is between the camera and object and full if it is at the object. In addition, landmark lines and corners are collected into a database for robot localization. The robot will continue to move into accessible open space collecting information about the environment through the stereo cameras until there is no more open space or some perimeter limit has been reached. Perimeter limits could be set by the length of a power cord on a vacuum cleaner or arbitrary maximum distances programmed before exploration has started. At each new location that the robot moves to and collects information, its pose or location is estimated using the current known 3-D grid and landmark information. This simultaneous localization and mapping (SLAM) uses recursive estimates of the pose to update the landmark and 3-D grid data for refinement.
|
