Glossary
What Is a Head-Mounted Camera in Robot Learning?
Last reviewed: October 2026
Short answer
A head-mounted camera is a camera worn on the head rather than held or fixed in the environment, so it moves with the wearer’s gaze and records the task the way the wearer’s own eyes see it: hands, tools, the target object, and the surrounding scene in one continuous egocentric view. In robot learning, this first-person viewpoint matters because it is close to the observation a robot with a head or chest camera actually occupies at inference time, unlike footage recorded from a fixed third-person angle or someone else’s point of view.
01
Why it matters
A policy trained on video from a viewpoint it will never occupy at inference time learns an observation distribution that does not match deployment. Egocentric video closes that gap by recording from the actor’s own point of view, because the observation distribution it produces matches what a robot with a head or chest camera actually sees. The head-mounted viewpoint is also what makes human demonstration transferable to a humanoid robot in the first place: a head-mounted camera occupies roughly the viewpoint a humanoid’s own head camera sees from, which is part of why humanoid programs lean on egocentric human video for pretraining-stage coverage. On its own, though, a head view has a structural weak point: as a hand approaches and grasps an object, the head camera’s view of the actual contact point is frequently occluded by the hand and arm, which is why most manipulation policies pair it with a wrist camera rather than relying on the head view alone.
02
How Firsthand handles it
Firsthand captures egocentric video from a head-mounted global-shutter 4K camera with a 120° field of view, hardware-triggered onto the same clock as the depth, IMU, and wrist-camera streams so every view is temporally aligned rather than stitched together after the fact. The head↔wrist↔depth transforms are solved against a shared calibration target and expressed in the head-camera frame, calibrated at both ends of every session, so the head view is geometrically related to the other streams, not just synchronized with them. Alongside the instrumented head-mounted rig, Firsthand also offers an Insta360-style head-mount capture tier for lighter-weight collection. Custom head-mounted camera collection runs against a written spec, is QA-checked against that spec, consented through Firsthand’s consent chain, and delivered under a perpetual, buyer-owned commercial license in RLDS, LeRobot format, or WebDataset.
03
Head-mounted camera vs. wrist camera: what each view is for
| Property | Head-mounted camera | Wrist camera |
|---|---|---|
| What it sees | The wearer’s wide egocentric view: environment, target object, and approach | A close, hand-frame view of the grasp point, resistant to the hand or arm occluding it |
| Best for | Scene context, navigation between sub-tasks, object localization | Contact detail, grasp precision, in-hand manipulation |
| Weak point alone | Loses grasp detail once the hand and object are close together | Loses scene context outside its narrow field of view |
| Firsthand delivery | Head-mounted global-shutter 4K camera, 120° field of view | Two wrist cameras (left and right) |
| Synchronization | Hardware-triggered onto the same clock as every other stream | Hardware-triggered onto the same clock as every other stream |
05
What a head-mounted camera stream is used for in training
| Stream | What it provides | Used for |
|---|---|---|
| Head-mounted video | The wearer’s wide egocentric view of the environment and task | Scene context, navigation between sub-tasks, object localization |
| Depth (head-frame) | Synchronized depth aligned to the head-camera frame | Scene and contact geometry, distance to target objects |
| IMU | Head motion and orientation on the same clock as video | Grounding the viewpoint’s motion for policies that reason about gaze and approach |
| Action segments | Labeled sub-task boundaries within the head-camera footage | Teaching a policy the structure of a task, not just raw frames |
06
Frequently asked questions
- What is a head-mounted camera?
- A head-mounted camera is a camera worn on the head that moves with the wearer’s gaze, recording a task from the wearer’s own first-person point of view rather than from a fixed third-person angle.
- Why does the head viewpoint matter for robot learning?
- A robot foundation model trained on third-person or internet video learns a viewpoint it will never occupy at inference time. A head-mounted camera records from close to the viewpoint a robot with a head or chest camera actually sees from, so the observation distribution matches deployment.
- Is a head-mounted camera enough on its own for manipulation data?
- No. As a hand approaches and grasps an object, a head-mounted camera’s view of the contact point is frequently blocked by the hand and arm. Most manipulation policies pair it with a wrist camera, which stays close to the object through that final phase.
- How do head-mounted and wrist cameras work together?
- The head camera provides the wide egocentric view of the environment, target object, and approach, while the wrist camera provides a close, occlusion-resistant view of the grasp point. Firsthand hardware-triggers both onto the same clock and solves their transforms against a shared calibration target, expressed in the head-camera frame.
- What makes head-mounted camera data usable for training rather than just footage?
- Calibration, synchronization, and labels. Firsthand’s head-mounted camera is hardware-triggered onto the same clock as depth, IMU, and wrist streams, calibrated against a shared target at both ends of every session, and QA-checked against the written spec before delivery.
- What capture setups use a head-mounted camera?
- Firsthand offers an instrumented head-mounted rig alongside an Insta360-style head-mount capture tier for lighter-weight collection, both recording from the wearer’s head-worn viewpoint.
- Does a head-mounted camera capture depth as well as video?
- Yes. Firsthand pairs the head-mounted camera with a synchronized depth stream aligned to the head-camera frame, in addition to video, IMU, and the wrist-camera streams.
- How do you get custom head-mounted camera data?
- Custom head-mounted camera collection runs against a written spec, is QA-checked against that spec, consented through Firsthand’s consent chain, and delivered under a buyer-owned commercial license.
Related terms