DATA OPERATIONS FOR PHYSICAL AI
Rodayta helps robotics teams turn real environments into structured training data. We plan capture, coordinate operators, annotate task events and validate deliveries against your specification.
BUILT AROUND YOUR SPEC
From camera placement to task taxonomy, every decision is connected to the model objective and the data contract. Start small, review representative sessions, then scale the workflow that passes.
01 / CAPABILITIES
Engage us for a focused workstream or an end-to-end program. The sensor stack, annotation schema and acceptance rules are agreed during scoping.
Task design and field execution for household, object-handling, packing and selected assembly workflows. We coordinate locations, operators, camera placement, SOPs and session logging.
Human-reviewed labels that describe what happened in the task. We define the ontology with your team, prepare a golden sample and keep edge cases visible during production.
Automated checks and human review catch technical and task-level failures before handoff. Issue logs, recapture decisions and accepted-session manifests keep each batch traceable.
02 / FIELD WORKFLOWS
We work from a task list and an explicit definition of a usable session. These are representative program directions. Final feasibility depends on the site, devices and acceptance criteria.
Kitchen routines, cleaning, sorting and object transfers captured as complete, observable task sequences.
Pick, inspect, pack and reset workflows with natural variation in items, positions and operator execution.
Feasibility-led pilots for repair, light assembly and tool handling, with safety and visibility requirements scoped first.
03 / ANNOTATION
A useful annotation workflow begins with clear definitions. We align the schema to your training objective, then review ambiguous events together before bulk labeling.
Concept illustration. Annotation classes and review thresholds follow the agreed specification.
Mark task boundaries, subtasks and meaningful action transitions with timestamped start and end points.
Describe manipulated objects, relevant state changes and the final condition needed to judge completion.
Separate completed attempts from partial or failed attempts. Record the cause when it is observable.
Escalate ambiguous footage, sample reviewer agreement and retain corrections alongside the batch record.
04 / ACCEPTANCE-LED QA
Each pilot starts with a written acceptance matrix. The review process combines file-level checks with human judgment about task visibility and completion.
Share your acceptance criteriaDevice settings, camera coverage, calibration evidence, task instructions and operator briefing.
Duration, resolution, frame rate, timestamp continuity, missing channels and synchronization where required.
Hand and object visibility, task progression, outcome, privacy issues and exception classification.
Every decision is tied to a reason code so the next batch can improve.
05 / WHAT YOU RECEIVE
The exact file structure is agreed in the pilot. This example shows how source media, annotations and QA evidence can travel together.
▾ accepted_batch/
▾ session_001/
◧ ego_video.mp4
◧ exo_video.mp4 if scoped
▤ timestamps.csv
▤ annotations.json
▤ accepted_sessions.csv
▤ issue_log.csv
▤ qa_summary.pdf
▤ schema_and_readme.md
We can organize the final batch around your session IDs, task taxonomy, media channels and metadata requirements. Format mapping and any training-framework conversion are scoped and tested on a golden sample.
Final modalities, formats and rights are defined in the statement of work.
06 / ENGAGEMENT MODEL
A small paid validation phase exposes camera, workflow and annotation issues early. Scope and commercial terms are agreed before collection begins.
Tell us the robot task, environments, sensors, target volume, output schema and acceptance rules.
Agree the capture setup, operator SOP, label taxonomy, review plan and delivery checklist.
Inspect representative sessions together, correct ambiguity and freeze the working specification.
Track usable data, issues and recapture decisions against agreed milestones.
07 / LET'S TALK DATA
Send a task specification or a short brief. We will use it to prepare a feasibility conversation and outline the information needed for a pilot.