What Humaid does

Humaid is a San Francisco-based, vertically integrated data collection platform that supplies real-world, egocentric demonstration data for Physical AI and robotics training. The company deploys trained human operators wearing calibrated multi-sensor rigs — egocentric stereo RGB-D cameras at 1920x1080/30fps, wrist-mounted cameras, 16-bit depth, ~30 Hz egocentric IMU plus ~200 Hz wrist IMU, 70-keypoint/18,439-vertex SMPL-family body tracking, and 21-joint HAMR v1 hand pose — alongside third-person and teleoperation capture, with all synchronized signals stored in MCAP containers. The output is model-ready datasets delivered in standard formats (HDF5, RLDS, LeRobot, or custom schemas), browsable through the company's web-based Robotics Data Explorer with 60+ metadata properties per sequence.

The company's go-to-market is enterprise B2B: direct sales through the website's contact and meeting-booking flow, on-site collection by trained operators in customer environments, and pipeline delivery to robotics teams. Primary verticals are manufacturing (bin picking, weld inspection, quality control, machine tending, palletizing) and warehouse/logistics (pick-and-place, inventory scanning, package handling), with adjacent verticals in construction, mining, hospitality, food & beverage, retail, and healthcare. A flagship exclusive pilot with AKFA Group — a Central Asian conglomerate with 55,000+ employees across 50+ companies in 12+ countries — provides access to hard-to-reach industrial data environments. The company employs 1–10 staff across San Francisco and Tashkent, Uzbekistan, with scientific direction led by Chief Science Officer Ruslan Salakhutdinov (former VP of AI Research at Meta, Professor at Carnegie Mellon).

The business model combines professional services (on-site data collection, annotation, QC) with data monetization (production-grade dataset delivery). Pricing is quote-based enterprise, with no public pricing disclosed. Revenue, funding rounds, and parent-company ownership are not disclosed in available sources. Co-founders Bobirjon Mardonov (CEO) and Doniyorbek Rakhmonberdiev (CTO) lead the operating team.

Name

Humaid

Legal name

Humaid

Website

https://humaid.co

Company type

Private

Founded year

2026

Operating status

Operating

Headcount range

1–10 employees

Short description

Humaid is a San Francisco-based Physical AI data collection platform that supplies enterprise robotics teams with real-world, egocentric, multi-sensor demonstration datasets collected by trained operators on-site, primarily serving manufacturing and warehouse/logistics customers.

Ownership category

akta.pro rank

Product category

Robotics Training Data

NAICS

Instruments and Related Products Manufacturing for Measuring, Displaying, and Controlling Industrial Process Variables (334513)

akta.pro primary industry

Industrial & Warehouse Robotics (HDAAAIAA)

akta.pro secondary industry

Construction & Mining Robotics (HDAAAIAE)

Keywords

• Robotics training data
• Human-in-the-loop data
• Egocentric video capture
• Physical AI datasets
• Teleoperation data collection

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices2 records

Markets served

GTM type

B2B

Offering type

Services

Cost components

Personnel, Operations, Technology or R&D, Infrastructure, Marketing or Sales

Revenue model

1. Data Collection Services: On-site real-world data collection services for robotics training. Includes calibrated multi-sensor capture, trained operators, annotation, QC, and pipeline delivery. B2B enterprise contracts for Physical AI training data.
2. Robot Training Data Delivery: Production-grade datasets collected by expert human operators in real environments. Delivered in standard formats (HDF5, RLDS, LeRobot, custom schemas) with full metadata and calibration files.

Go-to-market motion1 record

Distribution channels3 records

Marketing channels5 records

Core offering

Humaid collects real-world, egocentric demonstration data for robotics using calibrated multi-sensor wearable rigs and trained human operators across manufacturing, warehouse, hospitality, and food service environments. The company delivers on-site data collection, annotation, quality control, and pipeline delivery in standardized formats (HDF5, RLDS, LeRobot) for Physical AI imitation learning and robot foundation model training.

Product overview

Humaid is a vertically integrated data collection platform for Physical AI and robotics. The core offering consists of human-in-the-loop data collection services — including egocentric video capture, teleoperation recording, and multimodal sensor data — organized around a Robotics Data Collection Platform that coordinates trained operators, calibrated sensor hardware, annotation pipelines, and quality control into a standardized workflow. The platform delivers specialized vertical offerings for manufacturing, warehouse, and hospitality environments, with data accessible through an integrated Robotics Data Explorer web interface. All data is delivered in standardized formats (HDF5, RLDS, LeRobot) with full metadata for direct integration into robot training pipelines.

Differentiator

Problem solved

Functional benefit

Products and services

• Human-in-the-Loop Data Collection Real-world demonstration data collection service in which expert human operators perform physical tasks while calibrated multi-sensor rigs capture synchronized multimodal streams, including egocentric video, teleoperation recordings, and annotations, for robotics imitation learning.
• Egocentric Data Collection for Robotics First-person video and sensor data captured from the operator's perspective using head-mounted, wrist-mounted, or chest-mounted calibrated camera rigs, designed to match robot platform viewpoints for visuomotor policy learning.
• Robotics Data Collection Platform End-to-end platform spanning task execution by trained operators, multi-modal synchronized capture, annotation and quality control, and pipeline delivery in standardized formats (HDF5, RLDS, LeRobot) for robot training infrastructure.
• Robot Training Data Production-grade datasets including egocentric video, teleoperation recordings, force-torque data, and annotations for imitation learning and robot foundation model training, collected on-site in real operating environments and delivered for integration with training infrastructure.
• Manufacturing Robotics Data Real-world manufacturing data collection on production floors for tasks including bin picking, assembly, weld inspection, quality control, machine tending, palletizing, and material handling, delivered with object 6-DoF poses and contact force profiles.
• Warehouse Robotics Data Real-world warehouse and logistics data collection for pick-and-place, palletizing, inventory scanning, and package handling across thousands of SKU variations, delivered with grasp type, object category, weight class, and placement accuracy labels.
• Teleoperation Data Collection Human operators remotely control robot arms while recording joint positions, velocities, force-torque readings, and end-effector poses to produce action-labeled trajectories ready for behavior cloning and diffusion policy training.
• Physical AI Data Collection Real-world demonstration data, egocentric video, teleoperation recordings, and multimodal annotations that teach embodied AI systems to act in unstructured environments, closing the sim-to-real gap for Physical AI applications.
• Robotics Data Explorer Web-based interface for browsing, inspecting, validating, and downloading multimodal robotics datasets, featuring synchronized video playback with hand pose overlays, 60+ metadata properties per sequence, 3D visualization powered by Rerun.io, and downloadable data files in multiple formats.

Quantifiable outcome

• Closes sim-to-real gap by capturing real physics, real contacts, and natural edge case coverage
• +3 more outcomes

Named customers3 records

Segments7 records

Ideal customer profiles2 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

Integration1 record

AI capability9 records

Feature10 records

Partnerships

One partnership is on record.

• AKFA GroupflagshipStrategic or Co-development PartnerHumaid launched a groundbreaking exclusive pilot program with AKFA Group, one of the largest companies in Central Asia. The partnership enables data collection across AKFA's industrial operations including manufacturing (cement, steel, glass, building materials), supply chain & logistics, construction, mining, hospitality, food & beverage, retail, and healthcare. This provides access to environments with 55,000+ employees across 50+ companies in 12+ countries.

Scale indicators9 records

Recent moves6 records

Expansion highlights5 records

Direct peers

• Labelbox: AI data labeling platform with multimodal video, image, and sensor annotation tooling. Competes with Humaid's managed annotation and QC pipeline, especially for teams building foundation-model training datasets.
• Scale AI: Largest AI training data platform; expanded aggressively into robotics, autonomous systems, and government/defense data labeling. Directly competes with Humaid for physical AI demonstration data contracts with enterprise robotics teams.
• Sama: Managed data annotation company with on-the-ground operator networks for AI training data, including video and sensor streams. Direct competitor for human-in-the-loop collection services sold to enterprise AI and robotics teams.
• Encord: Data development platform for multimodal AI including video, image, and sensor fusion. Targets robotics and AV teams with annotation, curation, and evaluation workflows — overlapping directly with Humaid's data platform stack.
• Roboflow: Computer vision dataset and annotation platform widely used in robotics-adjacent teams. Competes for the same training data budget, though Humaid's differentiator is physical, on-site, multi-sensor capture rather than 2D vision datasets.

Broad incumbents

• Telus International (Lionbridge AI): Large-scale AI data services arm of Telus offering managed annotation, data collection, and validation across modalities. Competes with Humaid on enterprise data collection contracts where buyers prefer large incumbent vendors.
• Appen: Global incumbent in data annotation and collection for AI. While not robotics-specialized, Appen serves many of the same enterprise buyers with similar human-in-the-loop data services and represents the established alternative Humaid must displace.

Emerging players

• Snorkel AI: Programmatic data labeling and dataset curation platform for AI. Adjacent competitor focused on enterprise data development pipelines rather than raw collection, but competes for the same training data budget.
• Physical Intelligence: Robotics foundation model startup that has built its own large-scale teleoperation and demonstration data collection infrastructure. Represents the in-house alternative that Humaid's enterprise customers could choose instead of buying external data.
• Skild AI: Embodied AI / robotics foundation model company building large-scale robot training datasets internally. Comparable to Humaid as both operate in the robotics training data frontier, though Skild is a downstream model builder rather than a data supplier.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat4 records

Key risks5 records

Key highlights6 records

Customer concentration

• LinkedInlinkedin.com/company/humaidphysicalai

Revenue estimate

Valuation estimate

Number of profiles

Profiles7 records

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Humaid