• Human Action Dataset · Physical AI

Real human motion.
At industrial scale.

1M+
Hours Capacity
15,000
Custom Camera Rigs
10,000+
Factory Partnerships
100,000+
Household Partnerships
REC · 30FPS · 1920x1080 · 9DOF
01 · THE CHALLENGE

Humanoid robots need real-world human motion data — at scale.

THE GAP

Existing datasets lack first-person perspective and industrial diversity.

9 DOF IMU sensor-fused video is nearly impossible to source commercially.

Models trained on third-person or synthetic data fail to generalise.

OUR ANSWER

10,000 proprietary egocentric camera rigs deployed globally.

1 million+ hours of capacity deliverable within six months.

FHD 30FPS video fused with 9 DOF IMU sensors at every location.

02 · DATA SPECIFICATIONS

Built for training. Engineered for transfer.

VIDEO

ResolutionFHD 1920x1080
Aspect Ratio16:9
Frame Rate30 FPS
PerspectiveEgocentric

SENSOR

IMU9 DOF
AxesAccel • Gyro • Mag
SyncFrame-level
OutputCSV + Video

SCALE

Camera Rigs15,000 devices
Capacity1M+ hrs / 6mo
Factories10,000+ sites
Households100,000+ locations
03 · INDUSTRY COVERAGE

15+ verticals. Continuously expanding.

01Manufacturing
02Construction
03Retail
04Diamond Factory
05Footwear
06Household Work
07Hospitality
08Trades
09Diagnostic Labs
10Packaging
11Agriculture
12Waste Management
13Fulfillment Ops
14PVC Manufacturing
15Welding
04 · WHY PHYSICAL AI NEEDS THIS

From human demonstration to robot capability.

Use Case / 01

Humanoid Robot Training

Teach robots real-world tasks via authentic human demonstrations at industrial scale.

Use Case / 02

Embodied AI Foundation Models

Pre-train large action models on diverse egocentric video with IMU context.

Use Case / 03

Imitation Learning

First-person footage enables direct behaviour cloning from expert workers.

Use Case / 04

Task & Motion Planning

Richly labelled sequences for multi-step industrial and domestic workflows.

WHY EGOCENTRIC

  • Matches robot camera placement exactly
  • Captures hand-eye coordination natively
  • Preserves full spatial and depth context
  • Natural action segmentation from gaze
  • Directly transferable to robot POV
  • Rich attention & intention cues
05 · SAMPLE DATA

Evaluate quality before committing.

No sign-up required.

01
SAMPLE PACK 01

Industrial & Manufacturing

Apparel, electronics, electrical, footwear factory environments with synchronised IMU sensor data available on request.

Access on Drive ↗
02
SAMPLE PACK 02

Household & Domestic

Real home environments spanning diverse activities and demographics — daily tasks captured first-person.

Access on Drive ↗
LET'S BUILD IT TOGETHER

The future of Physical AI starts with real data.

Reach out directly to discuss capacity, custom verticals, licensing, or sample access.

founders@humanstryde.com →
Abhinaba Mondal · Founder, HumanStryde