REC · EGOCENTRIC · CAM_00 LACING · FOOTWEAR · 1080P60 28.4595° N · 77.0266° E — GURUGRAM, IN EST. 2026 · MANUDATA TECHNOLOGIES

TEACH ROBOTS
THE WAY HUMANS
actually work.

The world's largest untapped source of manipulation data — millions of skilled hands across India's MSME network. Real footage, hand-tracked, delivered RLDS-ready to frontier robotics labs.

Partner with us → Capture library
Feed select /
Scroll ▾
The thesis
While others strap cameras on a few hundred workers, we tap the most labour-intensive manufacturing economy on Earth — exactly where the hard manipulation tasks still live.
0MSMEs across India — every manipulation task imaginable
0Skilled workers performing contact-rich tasks daily
0Lower capture cost than US / EU programs
Capture index

SHOT ON THE
factory floor.

Full library →
21×33D keypoints / hand
60fps1080p egocentric
6DoFSLAM camera pose
nsTimestamp precision
Partnerships

LET'S BUILD THE
dataset TOGETHER.

partnerships@manudata.ai →
Capture library

ELEVEN TASKS.
real demonstrations.

Egocentric footage from live footwear, garment, and assembly lines — five clips carry 21-point hand-tracking overlays, the rest raw capture. Select any clip to play full-frame.

The full stack

RAW FOOTAGE →
robot-ready episodes.

A seven-stage pipeline runs every capture from camera to standardized training format — no manual annotation in the loop.

01

Capture

Cap-mounted egocentric 1080p60.

02

Track

SLAM recovers 6DoF camera path.

03

Hands

21-keypoint 3D pose, monocular.

04

Segment

Per-pixel object masks.

05

Localize

AprilTag world anchoring.

06

Annotate

Temporal action + language.

07

Export

RLDS / LeRobot, zero conversion.

Reconstruction

ONE CAMERA,
full 3D reconstruction.

This is a live replay of an actual capture from the sole-bonding line — both hands reconstructed in 3D from a single cap-mounted feed. WiLoR-estimated, MANO-derived joints, One-Euro smoothed, 99.9% detection across the take. The camera tracks the action automatically — drag to take control.

REPLAY — REAL CAPTURE WILOR · MANO-DERIVED · 21-KP ×2 STICKING · SOLE BONDING ⟲ Drag to orbit · f 000 / 000
episode · lacing_demo_001.rldsRLDS v1
// RLDS episode — one manipulation demonstration
{
  "episode": "lacing_demo_001",
  "steps": 3600,  // 60s @ 60fps
  "source": "msme_node_047",
  "step_t": {
    "rgb":        [1920,1080,3],
    "left_hand":  [21,3],   // 3D keypoints
    "right_hand": [21,3],
    "cam_pose":   [4,4],    // SLAM 6DoF
    "action":     "THREAD → PULL",
    "language":   "thread lace through eyelet, pull taut",
    "timestamp":  1748… // nanoseconds
  }
}
Field notes

FROM THE
factory floor.

Research, build logs, and what we're learning collecting manipulation data at scale.

ManuData
@manudata
ManuData Technologies
PHYSICAL AI · TRAINING DATA
Partnerships

TELL US WHAT YOUR
ROBOTS NEED TO learn.

Frontier lab, humanoid OEM, or data aggregator — describe the manipulation tasks you're training on. We'll scope a capture program and a sample delivery.

partnerships@manudata.ai ↗