Capture a robot workspace for later reconstruction.

Environment capture

Record a static workspace slowly and deliberately, with repeated viewpoints and stable scene conditions. These passes are source material for reconstruction and do not need to share the robot dataset timeline.

Synchronized robot-dataset recording

Record robot activity only when camera identity, timestamps, clock relationships, and drift are documented against the dataset. This supports time-aligned observation, not static scene coverage.

Phone video alone does not establish synchronization or alignment to the robot shared-world coordinate frame. A video can be useful in either workflow, but it is not evidence that their clocks or coordinate systems agree.

• Use a charged phone with enough storage and a clean rear lens.
• Confirm that original files and metadata can be transferred without messaging-app recompression.
• Prepare measured references, a pass log, and a safe walking route before recording.
• Keep emergency-stop access clear and account for cables, trip hazards, pinch points, and restricted areas.
• Use a spotter when the site procedure or visibility requires one.

Prefer a powered-down or safely inhibited robot for environment passes. Do not enter the operating envelope of a moving robot or trade safe separation for a camera angle. Recording robot motion requires the site's approved operating and synchronization procedure.

• Remove accidental clutter while preserving fixed geometry relevant to the workspace.
• Keep objects, chairs, cables, doors, and robot components stationary across passes.
• Add removable visual texture only when it is safe, non-damaging, and documented.
• Place measured scale references and calibration targets where several viewpoints can see them.
• Record every object that was covered, removed, moved, or added for capture.

Use one rear camera and lens consistently. Select a resolution and frame rate the phone can sustain, keep orientation fixed, and lock focus, exposure, and white balance when the device permits. Record the actual settings rather than assuming defaults.

Avoid digital zoom, lens switching, filters, portrait or cinematic effects, and modes that introduce changing geometry or exposure. Stabilization behavior varies by device and must be documented and checked rather than assumed beneficial. Preserve the original files.

1. Begin with a wide establishing pass showing the whole workspace and its fixed surroundings.
2. Walk a continuous perimeter with substantial overlap and gradual changes in viewpoint.
3. Translate around objects instead of standing in one place and only panning; reconstruction needs parallax.
4. Move slowly enough to avoid motion blur, abrupt turns, and skipped surfaces.
5. Make upper, middle, and lower passes, then targeted passes for safe-to-reach occluded areas.
6. Repeat critical geometry from more than one distance and direction without changing the scene.

If recording is interrupted, the lens changes, an object moves, or lighting changes materially, log the event and restart the affected pass. More footage is not a substitute for consistent, overlapping viewpoints.

Use stable, diffuse illumination. Avoid flicker, moving shadows, direct glare, and exposure changes between passes. Monitor screens, mirrors, polished metal, glass, transparent containers, repeating patterns, and featureless surfaces as likely failure regions.

When safe, cover or temporarily remove reflective or transparent objects and document the intervention. Otherwise expect holes, floaters, doubled surfaces, or unstable geometry. A visually convincing splat is not proof of geometric accuracy.

Record exact dimensions and units for multiple visible references where practical. Scale recovery is not frame alignment: a correctly sized reconstruction may still have the wrong translation, rotation, origin, or handedness.

Keep the robot in one documented, stationary pose during environment passes. Preserve its identity and joint-state or equivalent pose record. Capture fixed robot-base landmarks and calibration targets that can also be measured in the robot's frame.

Alignment to canonical-shared-world requires a documented calibration method, measured correspondences or an equivalent traceable procedure, a fitted transform, and validation against independent points. Seeing the robot in a phone video does not supply that transform.

Keep immutable originals separate from working files:

• originals/ — untouched phone files and hashes
• passes/ — pass inventory, route, purpose, and events
• calibration/ — target images, measurements, correspondences, and transforms
• reconstruction/ — tool versions, settings, logs, and output hashes
• validation/ — thresholds, measurements, overlays, decisions, and limitations

Record session ID, date, time zone, operator, location description, device and software versions, lens, resolution, frame rate, orientation, focus and exposure behavior, filenames, sizes, durations, hashes, lighting, scene changes, scale references, robot identity and pose source, safety state, privacy actions, and known failures. Do not call timestamps synchronized unless their clock relationship and drift were measured.

Record the reconstruction tool, version, settings, inputs, logs, and output hashes. Inspect tracking continuity, coverage, bounds, holes, floaters, doubled surfaces, and difficult materials. Compare reconstructed distances with measured references that were not all consumed by reconstruction or calibration.

Before fitting or judging alignment, predeclare task-specific acceptance thresholds and their rationale. Report transform residuals and validate distributed points that were not used to fit the transform. Retain visual overlays at robot bases and workspace landmarks alongside the numerical results. This guide does not prescribe universal distance or angle tolerances.

Blur or tracking loss

Slow down, improve stable lighting, and repeat the pass with smoother translation.

Holes or floaters

Add overlapping viewpoints and address occlusion, glare, transparency, or featureless regions.

Doubled geometry

Check for moved objects, people, shadows, exposure changes, and inconsistent passes; recapture a static scene.

Wrong scale or alignment

Recheck units, reference measurements, frame definitions, correspondences, and transform conventions.

Missing provenance

Recover originals and records if possible; otherwise document the gap and reject the real-scan claim.

Obtain permission to record the location and identifiable people. Exclude people, screens, badges, paperwork, addresses, network identifiers, and confidential equipment where possible. A reconstruction can preserve details that are easy to miss in one frame.

Prefer prevention during capture. If redaction is required, retain a record of what changed, apply the governing access and retention policy to originals, and verify that derived outputs do not reveal the removed information.

A manifest may use reconstructionClaim: "documented-real-scan" only when all of the following evidence is retained and reviewable:

• Immutable original capture inventory with file hashes, capture metadata, and pass log.
• Real-scene provenance and a record of every material scene intervention.
• Reconstruction inputs, tool and version, settings, logs, and output artifact hashes.
• Scale-reference measurements, units, uncertainty where known, and images showing placement.
• Robot identity, pose evidence, shared-world frame definitions, and transform conventions.
• Calibration method, raw measurements or correspondences, fitted transform, and provenance of the result.
• Predeclared task-specific thresholds, quantitative residuals, and independent validation points not used to fit the transform.
• Distributed visual overlays, known limitations, failed regions, and unresolved discrepancies.
• A traceable reviewer identity, review date, and accept or reject decision.

reconstructionClaim must remain false whenever required evidence is missing, contradictory, or outside predeclared tolerances. Passing this documentation gate does not by itself approve the asset for production rendering or robot control.

Capture and records

• All required surfaces have stable, overlapping coverage from multiple passes.
• Originals, hashes, settings, pass events, safety state, and scene changes are retained.
• Scale references, robot pose, and calibration targets are visible and measured.

Reconstruction and alignment

• Known failure regions and interventions are documented.
• Predeclared task-specific thresholds are recorded before acceptance is judged.
• Scale checks, calibration residuals, independent validation points, and overlays meet those thresholds.

Governance and claim

• Privacy, permission, access, and retention requirements are satisfied.
• A traceable reviewer identity, date, and decision are retained.
• The claim remains false unless every required evidence item is complete and consistent.
• No claim is based only on the existence of a phone video or visually convincing splat.