MEDIA ASSET PROCESSING & LABELING ENGINE

One capture.
Every dimension.

Turn real-world video into structured robotics data

Ordering on GRID? Add MAPLE to any order.

Recorded dishwashing result with object boxes, hand skeletons and wrist trails
Recorded result from a real 152-second dishwashing capture.
MAPLE in 15 seconds: real-world video in, every frame inspectable, video, motion, depth and 3D in sync, and what you get.

How it fits.

MAPLE gets a conforming hour ready for your training pipeline.

  1. Capture

    Partners film to your spec in the Waffle app, with video and sensors captured together.

  2. Check

    Automated checks run on every clip against the order's spec.

  3. MAPLE

    Labels, privacy masks, spatial data and export packs, tuned to your spec.

  4. Deliver

    Conforming hours land in your pipeline, in the formats you train on.

Already have footage? MAPLE runs on video you bring, too.

Every frame, inspectable.

Objects, hands, actions, scenes and privacy, labeled and reviewable.

Technical details

Scores are uncalibrated. All boxes remain in the sidecar.

Hand poses are estimates, not verified metric world-space ground truth.

Per frame: objects, hands, actions, scene labels, quality metrics and 21-joint hand estimates.

The capture, in every dimension.

Video, motion, depth and 3D, in sync.

Recorded depth heatmap
4,565camera poses
15,128IMU samples
1,277depth maps
97mesh snapshots

From this recorded capture. Spatial data is included when your captures carry it.

Technical details

See what a model receives.

Training-ready samples. Uncertainty kept in.

Technical details

Exports contain human demonstrations, not robot actuator commands.

4,564 of 4,564 rows read by the official LeRobot loader, on the recorded dishwashing capture.

Privacy, handled.

  • Faces and text are detected on every frame.

  • An operator reviews the detections.

  • A blur or solid mask is baked into the export.

What you get.

.json

Sidecar JSON

Every label, one file per clip.

.mp4

Reviewed video

Your privacy decisions, applied.

LeRobot v3

Human demonstrations

Packaged for your training pipeline.

WebDataset

Training pack

Shards ready to stream into training.

Tables

Annotation tables

Every label, ready to query.

3D

Spatial data

Depth, scene mesh, point cloud and camera trajectory, when your capture includes them.

Labels, checks and privacy screening, tuned to each buyer's spec.

Scoped with you, project by project.

Tell us what you're training, and we'll work out the scope and the price together.

Talk to sales

Ordering on GRID?

Add MAPLE to any order.

Browse GRID

Bring your footage.

Tell us what you have and what you train on. We'll send an estimate.

Talk to our team
What video and sensor inputs do you support?

Egocentric and fixed-camera video, including iPhone sensor captures. Captures from the Waffle app arrive with video and sensors already fused. When supplied with your captures, MAPLE can also inspect IMU, depth/LiDAR, recorded camera poses, meshes and MCAP recordings. Tell us what you have and we will confirm fit.

Are the exports robot commands?

No. MAPLE packages human demonstrations: annotated video, estimated hand poses, object labels and task labels in JSON and LeRobot format. They contain no robot actuator commands, and hand poses are estimates rather than calibrated world-space ground truth.

How do I request a quote?

Email sales@wafflevideo.ai with a short description of your footage (volume, capture setup, sensors) and the outputs you need. We will send an estimate before any work starts. There are no public uploads or self-serve checkout.

Can I add MAPLE to a GRID order?

Yes. Add MAPLE to any order. It's optional, and we scope the work and the price with you for each project.