Training-ready data for AI and robotics.
Annotyx AI collects, annotates, validates, and delivers project-specific datasets for computer vision, robotics, embodied AI, and multimodal models.
Send us existing data or commission a new real-world dataset with custom tasks, environments, participants, camera perspectives, annotation ontologies, quality controls, and structured delivery.
Pilot to production scale · Multiview capture · Human QA · Privacy review · Structured exports
FRONT CAM // 1080PDEPTH MAP: READY
SIDE CAM // 1080P{"episode":"AAI-KITCHEN-004281","action":"cut","object_state":"sliced","outcome":"success"}Two Core Services
One platform for annotation and real-world data production.
Annotyx AI operates the full data workflow: ingest existing source data or commission a new project-specific collection, then move from ontology design through quality control to structured delivery.
Your data, labeled and training-ready.
Supply images, video, documents, text, audio, 3D, LiDAR, or sensor data. Our platform supports project-specific ontology design, human review, validation, format conversion, and structured delivery.
New physical-world data, built for your model.
You define the activities, scale, perspectives, participants, environments, sensors, and labels. We design the collection program, operate contributor workflows, synchronize recordings, annotate, validate, and deliver.
From focused pilots to multi-environment collection programs.
Need 100, 1,000, 10,000, or significantly more hours of task-specific demonstrations? Annotyx AI designs the collection protocol, contributor workflow, quality gates, annotation schema, and delivery pipeline around your requirements.
Scale across tasks, participants, environments, viewpoints, sensor configurations, and annotation depths. Collection programs can include thousands of recording sessions and large volumes of synchronized video.
Distributed Operations
Structured contributor assignments, qualification, equipment specifications, capture protocols, and project tracking.
Configurable Depth
Scale from raw capture and metadata to temporal actions, object states, interactions, segmentation, pose, and multimodal labels.
Controlled Delivery
Privacy review, human QA, version control, usage-rights records, and delivery status remain visible through the control workflow.
Target Domains
Who Annotyx AI serves.
We support teams building computer vision, generative AI, automation, robotics, and embodied intelligence systems.
Computer Vision Teams
Bounding boxes, segmentation masks, keypoints, classifications, and object tracking.
LLM & Generative AI
Text classification, response evaluation, preference ranking, and RLHF-ready labels.
Audio & Speech AI
Transcription, speaker labels, intent, events, and quality review for audio datasets.
Humanoid Robotics
Demonstrations tailored to dual-arm manipulation and whole-body locomotion pretraining.
Service Robotics
Commercial facility routines, tidying, and tool deployment sequences.
Household Robotics
Domestic manipulation tasks inside controlled mock apartment environments.
Cleaning Robotics
Surface wiping, spill management, and vacuum routines under varied angles.
Warehouse & Logistics
Package handling, parcel sorting, and shelving tasks in mock storage hubs.
Foundation Models & Embodied AI
General-purpose dataset structures designed to feed Vision-Language-Action (VLA) architectures.
World Model Research
Multiview environment and physics sequences for structural world prediction training.
AI Data Companies
Annotation, synthetic augmentation, and consent-clean data to diversify broad catalogs.
Universities & Research Labs
Academic-aligned custom pilots to validate imitation learning and manipulation theories.
Physical AI models cannot scale on fragmented, synthetic, or scraped video logs.
Many teams train policies on unstandardized datasets that suffer from severe limitations:
- Legal & Ethical liabilities: Web-scraped clips lack explicitly documented, commercial-use contributor consent.
- No annotation depth: Raw video files miss temporal action boundaries, grasp-contact triggers, and object states.
- Environmental bias: Fixed camera angles and single-view logs fail to generalize across real-world workspaces.
Rigorous, consent-documented human demonstration workflows.
We design custom protocols that isolate physical primitives and output structured, release-ready dataset packages:
- Multiview recording: Synchronized ego-worn (chest/head) and static environmental views capture the full workspace state.
- Consent and usage rights: Required permissions, participant consent, access rules, retention, and delivery rights are scoped for each collection.
- Complete annotation layer: Level 1 metadata down to Level 7 outcome and recovery logs, formatted in parquet and JSONL.
Simulation Gap & Occlusion Blindspots
Missing grasp dynamics, noisy labels, and unverifiable commercial consent
Capabilities Preview
Dataset products for physical-world intelligence.
Versioned, consent-documented and quality-reviewed task demonstration datasets with multiview media, structured annotations, metadata, and defined licensing.
Service Work Demonstration Dataset v0.1
The flagship Annotyx AI collection: consent-based human task demonstrations spanning cleaning, tidying, object handling, and light logistics in controlled environments.
Annotation layers
Cleaning & Surface Interaction Dataset
Focused demonstrations of surface wiping, spill response, and simulated floor-cleaning routines across mock service environments.
Annotation layers
Household Tidying Mock-Apartment Dataset
Tidying, folding, and organizing demonstrations captured in a furnished mock apartment built specifically for controlled data collection.
Annotation layers
Operating Pipeline
From protocol design to validated delivery.
A controlled operating system from model requirements and contributor onboarding through synchronized capture, annotation, validation, and versioned delivery.
PHASE_01
Requirements
Model objectives, tasks, target volume, environments, viewpoints, sensors, and delivery formats are defined.
PHASE_02
Ontology Design
Classes, actions, states, temporal boundaries, edge cases, and acceptance criteria become one production schema.
PHASE_03
Capture Planning
Recording protocols align equipment, camera placement, variations, privacy rules, and technical specifications.
PHASE_04
Contributor Onboarding
Selected contributors complete identity and age verification, consent, training, and project qualification.
PHASE_05
Data Collection
Project-specific sessions are assigned, recorded, uploaded, tracked, and scaled across required environments.
PHASE_06
Privacy Review
Consent scope, faces, screens, identifiers, and sensitive details are reviewed before release processing.
PHASE_07
Raw Data QA
Media is checked for protocol compliance, completeness, visibility, synchronization, and technical integrity.
PHASE_08
Synchronization
Multiview streams, timestamps, sensor channels, metadata, and episode identifiers are normalized and aligned.
PHASE_09
Annotation
Objects, actions, states, interactions, outcomes, captions, and project-specific labels are applied.
PHASE_10
Validation
Human review, consensus checks, and automated schema validation protect consistency at production scale.
PHASE_11
Structured Delivery
Versioned media, labels, metadata, rights documentation, and QA reports are exported to the required structure.
Metadata & Quality Assurance
Structured annotations, not raw footage dumps.
Project workflows can include redaction, labeling, manual review, consistency checks, and schema validation before delivery.
- Task verb ontology and natural language instructions
- Precise temporal segments (start/end millisecond timestamps)
- Workspace environment, camera specification, and active object labels
- Privacy clearing blurs applied on screens, badges, and faces
- Quality grades (A · B · C · Reject) with structured check logs
Annotation Level 2 Preview
{
"episode_id": "AAI-SORT-000042",
"start_time_s": 4.20,
"end_time_s": 6.85,
"action": "pick_up",
"active_object": "plastic_bottle",
"active_hand": "right",
"outcome": "success",
"confidence": 0.97
}Record safe workspace tasks. Get paid for accepted clips.
Annotyx AI operates structured contributor programs for consent-documented, project-specific real-world data collection. B2B collection campaigns and contributor assignments remain separated through controlled access, training, review, and acceptance workflows.
Application, identity and age verification.
Consent, project training and qualification.
Task assignment and equipment requirements.
Protocol-led capture and secure upload.
Privacy and quality review, acceptance or repeat, then payment.
Minimum Age Required
18+ Years Old
Equipment Requirements
Defined Per Project
Compensation Structure
Per Approved Episode
Project Availability
Location Dependent
Need existing data labeled or a new training dataset?
Describe the data type, activities, volume, environments, perspectives, annotations, and target format. We will design the pilot, production workflow, quality gates, and structured delivery around your requirements.