Head Camera

A camera mounted on a robot's head or upper body that gives a near-first-person view of the scene.

A head camera is mounted on a humanoid's head, or above a mobile manipulation platform, giving a near-first-person view similar to looking out through one's own eyes. It's usually paired with a wrist camera and a fixed third-person camera: the head camera sees the overall scene and both hands, while the wrist camera sees close-up detail near the gripper. Mobile ALOHA, for instance, uses two wrist cameras plus one forward-facing camera mounted up top; Unitree's G1 carries a depth camera and a 3D lidar on its head. When the head can rotate, the camera's extrinsics change with the neck joint and must be recomputed in real time through kinematics. Teleoperation setups also use an ‘active head camera’: Open-TeleVision mounts a ZED Mini stereo camera on a 2-DOF gimbal that follows the operator's head rotation inside an Apple Vision Pro headset and streams back a stereo view. When collecting human data, a camera worn on a person's head or on glasses is likewise called head-mounted, capturing first-person video.

ExampleTraining a policy for dual-arm clothes-folding feeds in three camera streams: the head camera sees the whole garment laid out flat, while the two wrist cameras check whether the gripper has hold of the garment's corner.

Also called

Head-Mounted Camera, Top Camera

Related

Wrist Camera · Third-Person Camera · Egocentric Video · Open-TeleVision · Multi-View · Camera Extrinsics

Sources

Mobile ALOHA (arXiv:2401.02117)
Open-TeleVision (arXiv:2407.01512)
Unitree G1 产品页 (Chinese)