Egocentric vision for robotics with 4 wide-angle environment cameras, 2 downward-facing cameras, and datasets linking human gaze with surrounding scenes.