10
CAPTURE INDUSTRIES
21
OUTPUT FORMATS
4
LICENSING PATHS
12
ANNOTATION TYPES
Real-World AI Data
Captured to Specification
Built for Any AI Use Case
Fearless vision, delivered through real-world capture, multimodal annotation, and flexible licensing.
Human-CapturedAI TrainingData ForFrontier AI
delovantage/ sample data
See our data in action
[
{ "start": 0.0, "end": 2.0, "text": "A person is assembling electronic components on a workbench, carefully handling small parts and tools." },
{ "start": 10.0, "end": 12.0, "text": "A person is holding and examining a small electronic component at a workbench with various tools and parts around them." },
{ "start": 20.0, "end": 22.0, "text": "A person wearing a red shirt and a white apron with a cartoon character on it is holding a small object, possibly assembling or inspecting it." },
{ "start": 30.0, "end": 32.0, "text": "A person is holding a white object with a pink and black design, possibly a piece of clothing or fabric, in front of a green shelf filled with various items." },
{ "start": 40.0, "end": 42.0, "text": "A person wearing a red shirt and a white apron with a cartoon character on it is holding a small object, possibly a battery or a similar item, over a green table filled with various items." },
{ "start": 50.0, "end": 52.0, "text": "A person wearing a red shirt and an apron with a cartoon character is working on assembling or repairing electronic components at a workbench." },
{ "start": 60.0, "end": 62.0, "text": "A person is holding a green object, possibly a piece of equipment or a tool, while another individual in the background appears to be working on something with their hands." },
{ "start": 70.0, "end": 72.0, "text": "A person is sorting through small, colorful items on a table, possibly assembling or organizing them." },
{ "start": 80.0, "end": 82.0, "text": "A person wearing a red sweater with a cartoon character on it is reaching up to grab something from a shelf." },
{ "start": 90.0, "end": 92.0, "text": "A person wearing a red shirt and an owl-patterned apron is holding a small electronic component, possibly a circuit board or a module, with both hands." },
{ "start": 100.0, "end": 102.0, "text": "A person wearing a white glove is holding a small electronic component over a workbench with various tools and parts." },
{ "start": 110.0, "end": 112.0, "text": "A person wearing an owl-patterned shirt is holding and examining a small object, possibly a phone or camera, with both hands." },
{ "start": 120.0, "end": 122.0, "text": "A person wearing a red shirt and an owl-patterned apron is handling a large, shiny, metallic object on a green surface." },
{ "start": 130.0, "end": 132.0, "text": "A person is holding and examining a small, colorful object, possibly a component or part, in front of a workbench with various tools and materials." },
{ "start": 140.0, "end": 142.0, "text": "A person is holding and examining a small, rectangular object with a blue label on it, surrounded by various electronic components and packaging." },
{ "start": 150.0, "end": 152.0, "text": "A person wearing a white apron with a cartoon character on it is handling small, clear plastic bags filled with items, possibly for packaging or organizing." },
{ "start": 160.0, "end": 162.0, "text": "A person wearing a red shirt and a white hat with a cartoon character on it is reaching up to grab something from a shelf." },
{ "start": 170.0, "end": 172.0, "text": "A person is holding and examining a small electronic component, possibly a circuit board or module, with various other components visible on the table behind them." }
]delovantage / how it works
From real-world capture to your model, licensed end to end.
[ 02 ]Explore Our Data
Discover the industries powered by real-world first-person AI training data.
Our capture programs are scoped around real tasks and environments, with requested annotations, delivery formats, participant requirements, and licensing terms documented for each engagement.
Curated from real-world environments & expert annotation.
We don't just collect data
We build AI intelligence
✦ ✦ ✦
Logistics & Delivery
Shelf stocking, warehouse picking and packing, and customer interaction data from real retail workers.
Building AI from the real world
Output Formats
Requested datasets can be delivered in standard formats selected for the target pipeline. Open any card to review the format.
--
21 industry-standard output formats
MCAP Recording
H.264 Egocentric Video
Depth Maps
6-DoF Pose
Hand Mesh (MANO)
Hand Skeleton 2D/3D
Point Cloud (LIDAR)
Object Detections
Action Labels
SMPL Body Mesh
Tactile Sensor Data
ROS2 Bag
Sensor Telemetry CSV
Gaze Tracking
Stereo 180° Video
Scene Graph
Semantic Segmentation
Optical Flow
3D Object Tracking
Wrist Trajectory Prediction
Face Landmarks
Robotic Simulation Data
Physics-based sim environments, synthetic egocentric datasets, and sim-to-real transfer data.
On-premises Data Annotation
Configure annotation processing inside a customer-controlled VPC when project security requirements call for on-premises deployment.
Request a Custom Format
Can't find what you need? Tell us what you're building and we'll confirm whether the required delivery format can be included in the project scope.
Four ways we work with you
Flexible licensing for every stage of AI development — from research to production.
Exclusive License
Only your company accesses this dataset. No competitor gets the same captures. Built for teams where data uniqueness is the edge.
Non-Exclusive
Shared access to the same high-quality datasets. A cost-effective path for benchmarking and baseline model training.
Catalog Datasets
Review existing capture categories and available annotations, then define delivery formats and usage rights for the project.
Custom Program
Fully bespoke capture and annotation scope. Task definitions, environments, sensors, and delivery format designed around your model's exact requirements.
Data Preview Library
First-person POV footage captured across industries. Full previews coming soon.
6 categories
Urban intersection navigation
Driver-view capturing lane changes, pedestrian interaction, and real traffic patterns at complex intersections.
Commercial kitchen prep workflow
First-person kitchen operations — ingredient prep, food handling, plating sequences in a live commercial environment.
Vehicle assembly line operations
Worker-view on the assembly floor — component fitting, torque checks, and quality inspection workflows.
Precision component assembly
Egocentric footage of skilled workers performing sub-millimeter assembly and inspection tasks on the factory floor.
Shelf stocking & warehouse picking
Worker POV across shelf replenishment cycles, warehouse picking routes, and customer interaction at point of sale.
Last-mile delivery sequence
Full delivery run from van exit to handoff — curb approach, package scanning, doorstep delivery, and multi-stop routing.
Get the dataset
your model needs.
Tell us what your AI needs to see. We'll match it to something in the catalog or build a custom capture program.
Physical AI data FAQ
Questions AI teams ask before licensing data
What is Physical AI training data?
AI training data from Delovantage is real-world, human-captured multimodal data used to train foundation models, generative AI systems, vision models, embodied AI, and any system that needs to understand or replicate human behaviour. It can include first-person POV and external video, depth, pose, actions, hand-object interactions, telemetry, point clouds, tactile signals, and other synchronized sensor data.
Which industries does Delovantage AI Labs cover?
The current catalog focuses on automotive, manufacturing, retail, food and beverage, and logistics environments. Custom capture programs can scope additional tasks and environments when an existing dataset does not match the model requirement.
What output formats are available?
The current catalog lists 21 available formats, including H.264 video, MCAP, depth maps, 6-DoF pose, MANO hand mesh, detections, action labels, point clouds, ROS2 bags, gaze, semantic segmentation, optical flow, tracking, trajectories, and scene graphs.
What is the difference between exclusive and non-exclusive licensing?
An exclusive license restricts a defined dataset to one customer under the agreed scope. A non-exclusive license allows the same dataset to be licensed to multiple customers and is commonly used for benchmarking, research, and baseline model development.
Can Delovantage run a custom data capture program?
Yes. A custom program begins with the target tasks, environment, participant requirements, capture hardware, annotation modalities, quality criteria, delivery format, and licensing scope required by the AI team.
Need a task, modality, or environment not listed here? Scope a custom capture program.
[ 07 ]What we are building toward
Data is where we start, not where we stop.
- AvailableReal-World AI Data
- 21 delivery formats, 10 capture industries, 12 annotation types, 4 licensing paths — all listed on the home page with sample output.
- ResearchRobotics Intelligence
- Which layer of robot autonomy does our data advantage unlock first — perception, world modelling, or manipulation policy?
- ExploringIntelligent Devices
- Would a purpose-built wearable capture rig produce materially better training data than the commercial hardware we use today?
Only one of these is a product today. The other two are published as open questions, with the evidence policy in full on the research page.
What's new at Delovantage.
See more on the blogStart licensing real-world
AI data with Delovantage
Pick a dataset from the catalog, or describe a scenario we haven't captured yet — we'll build a custom program for it.