EyeRobot 2.0: Active Gaze for Precise Manipulation without Wrist Cameras
EyeRobot 2.0 uses active stereo gaze to recover much of the manipulation performance normally supplied by wrist cameras.
The system swivels two eye viewpoints toward 3D fixation points, then spends more visual computation on the image centers. Its gaze controller and target selector are trained on real-world data, with the selector learning fixation choices alongside the gripper policy. In the paper’s trials, passive stereo success fell to 27% without wrist cameras, while EyeRobot 2.0 outperformed passive stereo by 40% in real-world tasks. It also beat ego-plus-wrist policies when wrist views were blocked by grasped objects, 48% to 22%. HF Daily Papers' note
The system swivels two eye viewpoints toward 3D fixation points, then spends more visual computation on the image centers. Its gaze controller and target selector are trained on real-world data, with the selector learning fixation choices alongside the gripper policy. In the paper’s trials, passive stereo success fell to 27% without wrist cameras, while EyeRobot 2.0 outperformed passive stereo by 40% in real-world tasks. It also beat ego-plus-wrist policies when wrist views were blocked by grasped objects, 48% to 22%. HF Daily Papers' note
score 5