Connect and synchronize
Device setup, raw records, clocks and calibration.
Data operations
Data Platform
Turn collection records into reusable datasets. Connect devices, align time, clean and annotate recordings, review quality and publish versions for research and enterprise data delivery.
Process and review
Cleaning, segmentation, annotation and exception review.
Release a dataset
Versions, permissions, evaluation splits and export manifests.
Train and evaluate
Environments, weights, metrics and physical robot logs.
Learn from failures
Identify missing scenarios and collect corrective demonstrations.
Data dashboard
From one recording to the whole project.
Connect collection progress, task coverage and reviewed samples in one project view.
- Data overview
- Track collection volume and quality review outcomes.
- Task coverage
- Organize episodes by task, scene and modality.
- Sample browser
- Find reviewed samples by scene, task and annotation.
Tasks & scenesShared project context
Visual recordsInertial signalsTask labels
Quality reviewSample browser
Conceptual workflow. Access and available records depend on the project.
Field collection
Connect tasks with the people collecting them.
Define the task workflow and deployment scope with your project team.
Define a task
Specify the goal, scene, equipment and valid-sample requirements.
Publish and assign
Set participation, collection instructions and submission rules.
Track submissions
Review progress and records associated with the task.
Review and recollect
Turn quality feedback into corrections or additional collection.
Keep observations and conclusions distinct.
Link each episode to its task, device, time and calibration. Derived results preserve their source and version without replacing the original recording.
Raw observations
Video, FMG, IMU, tactile signals and robot logs.
Calibration and estimates
Clock alignment, coordinate transforms and estimated hand or tool poses.
Annotations and labels
Task segments, events, human review and robot action labels.
Dataset releases
Training, validation and test splits, export manifests and licenses.
Agree on the data contract.
Identity and versions
Project, task, episode and device IDs; firmware, calibration, algorithms and dataset versions.
Time and space
Device and reference clocks, synchronization method and errors, coordinates, units, extrinsics and validity intervals.
Quality and provenance
Dropped frames, missing data, exceptions and review results; identify measurements, estimates and human labels.
Rights and evaluation
Collection consent, usage scope and access control; separate evaluation data by participant, scene or date.
This is a project data-contract framework. Final fields, export formats and platform functions depend on the selected version and sample validation.
Scale collection with measurable acceptance.
Use a pilot batch to establish quality and processing cost before planning people, devices, scenes and delivery batches.
Every delivery batch
- Raw and processed data with a field dictionary and reading instructions.
- File manifests, versions, checksums and calibration files.
- Accepted-sample counts, coverage distributions and quality reports.
- Rejected-sample reasons, rework records and usage-rights documentation.
Before collection starts
- Define accepted episodes, duration and completed-task samples.
- Set timing, missing-data and annotation-consistency thresholds.
- Agree on task, object, environment, participant and failure coverage.
- Confirm ownership, U.S. storage, retention and third-party processing permissions.
Use failures to plan the next batch.
Trace failure states to recollection requirements. Release corrective demonstrations as a separate version and evaluate against a fixed test set, while retaining broad scenario coverage.
Build with Actrium
Start with your task.
Define the devices, data and evaluation your project needs.