What it does ¶
ALVisualSpaceHistory keeps track of the head movements of the robot to build a timestamped map of the head positions. This can be useful when exploring the environment to make the robot look in every direction.
How it works ¶
ALVisualSpaceHistory is an ALExtractor that regularly queries the head angles of the robot and updates a grid containing the timestamp of the last time the head looked in a particular direction.
- Visual grid data
- The “visual space” denotes all the possible head positions of the robot. If we took snapshots of what the camera sees at different locations and stitched them together, we would obtain a “visual map”. ALVisualSpaceHistory does not store such a map but creates a grid by dividing it into small rectangular cells which contain the timestamp of the last time that the robot’s camera could see this area of the visual space.
- Grid precision
The cell size is determined by the parameter grid precision which can be changed by calling ALVisualSpaceHistoryProxy::setGridPrecision().
This precision is the ratio of the camera field of view in radian divided by the field of view corresponding to one grid cell. If the precision is 2.0, then a grid cell represents a quarter of the camera field of view. A high grid precision gives a more accurate grid but also requires more processing time.
On every grid update the module pushes data to the ALMemory key VisualSpaceHistory/VisualGrid/Data.
Performances and limitations ¶
If the robot moves more than its head, the grid must be manually reset by calling ALVisualSpaceHistoryProxy::resetGrid().
Indeed, the “visual grid” is build using only the head position, it becomes obsolete as soon as the FRAME_TORSO space moves with respect to the FRAME_WORLD (in case of displacement on the ground for example). For further details, see: Frames.