What xdof does

XDOF (xdof, Inc.) is a robotics data infrastructure company building the data layer for physical AI and robot foundation models. The company was founded by Philipp Wu, Yide Shentu, and Nemo Jin — researchers from UC Berkeley who authored GELLO, a teleoperation system now widely deployed for robot data collection — and emerged from stealth in June 2026 with a $70 million seed round led by Thrive Capital alongside Spark Capital, Andreessen Horowitz, Lux Capital, and WndrCo. XDOF serves frontier AI labs and robotics companies developing general-purpose robots, operating as an embedded full-stack partner that owns teleoperation data collection, annotation, evaluation, and delivery on behalf of customers through Forward Deployed Engineers.

The company's technology stack spans teleoperation-based data collection systems, multimodal sensor capture (cameras, IMU, data gloves, force sensors), automated annotation pipelines with pose estimation and visual SLAM, and an exabyte-scale data warehouse. Its principal public artifacts include the ABC-130K open-source bimanual manipulation dataset (130,000+ trajectories across 195 tasks using I2RT's YAM arms, plus ~400 hours of simulation and evaluation data, released under Apache 2.0), the SARM2 stage-aware reward model (mixture-of-experts across 21 action primitives) with the SPIRAL autonomous self-practice framework, and the RA-BC reward-aligned behavior cloning training method.

XDOF monetizes through managed B2B services, charging customers for data collection, annotation, and evaluation delivered against custom specifications; pricing is quote-based and not publicly disclosed. The company has approximately 60 employees, ~20 active customers including leading frontier AI labs, and offices in San Mateo (headquarters), Shenzhen, and Jakarta.

Name

xdof

Legal name

xdof, Inc.

Website

https://xdof.ai

Company type

Private

Founded year

2026

Operating status

Operating

Headcount range

11–50 employees

Short description

XDOF (xdof, Inc.) is a robotics data infrastructure company providing teleoperation-based data collection, annotation, and evaluation services to frontier AI labs and robotics companies developing robot foundation models, augmented by proprietary reward-modeling research.

Ownership category

akta.pro rank

Product category

Robotics Data Infrastructure

NAICS

Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (518210), Computer Systems Design and Related Services (54151), Other Computer Related Services (541519)

SIC

Services-Computer Processing & Data Preparation (7374), Services-Computer Integrated Systems Design (7373), Services-Prepackaged Software (7372)

akta.pro primary industry

Domain-Specific Training Data Solutions (e.g., medical, AV, geospatial) (HDAAALAL)

akta.pro secondary industry

End-to-End Enterprise AI Platforms (MLOps & Model Lifecycle Management) (HDAEANAA)

Keywords

• Robot training data
• Physical AI infrastructure
• Teleoperation datasets
• Robot manipulation data
• Data annotation services

Headquarters

HQ city

Berkeley

HQ country

United States

HQ region

North America

Offices4 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

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

Revenue model

1. Robot Training Data Platform Services: XDOF provides data pipelines, collection tools, and annotation systems to help AI labs and robotics firms develop training data for physical AI. The company charges for data collection, annotation services, and evaluation services for robotics training.

Go-to-market motion2 records

Distribution channels2 records

Marketing channels3 records

Core offering

XDOF develops core robotics technology and data collection infrastructure for the world's largest AI research labs. The company provides end-to-end data pipelines, teleoperation data collection systems, annotation services, and an exabyte-scale data warehouse to produce high-quality training data for physical AI and robot foundation models. Beyond managed data services, XDOF ships proprietary reward modeling (SARM2) and autonomous self-improvement (SPIRAL) software, plus the open-source ABC-130K bimanual manipulation dataset.

Product overview

XDOF is a robotics data infrastructure company offering a unified platform for robot foundation model development. The core offering consists of data collection systems, exabyte-scale data warehousing, and annotation pipelines that enable AI labs to obtain high-quality training data for physical AI. The product portfolio includes ABC-130K (the largest open-source teleoperation dataset), SARM2 (a general reward modeling system for robot grading), and the SPIRAL autonomous practice framework. XDOF outsources operational robot fleet management so AI labs can focus on model development while XDOF handles data collection, processing, and quality control.

Differentiator

Problem solved

Functional benefit

Products and services

• XDOF Robotics Data Infrastructure Platform End-to-end managed platform providing data pipelines, teleoperation data collection, annotation systems, and exabyte-scale data warehousing for AI labs and robotics companies building physical AI.
• ABC-130K Dataset Open-source teleoperation dataset of 130,000+ bimanual robot manipulation episodes across 195 tasks with simulation and evaluation data, released under Apache 2.0 license for AI research labs.
• SARM2 Reward Modeling System Stage-Aware Reward Model 2 — a general reward modeling system using a mixture-of-experts architecture with 21 action primitives that grades robot task performance across diverse manipulation tasks to enable autonomous self-improvement.
• SPIRAL Autonomous Practice Framework Autonomous practice framework that lets robots attempt tasks, grade their own progress with SARM2, and refine behavior through reinforcement learning with minimal human intervention.
• Data Collection Systems Scalable teleoperation and egocentric data collection platforms with multimodal sensor capture, calibration, and validation tooling that produce high-quality robot training data for AI labs.
• Annotation Pipelines Automated annotation systems that convert raw multimodal sensor data into structured training labels for vision-language-action model training, including cross-modal temporal alignment and quality filtering.
• Exabyte-Scale Data Warehouse Large-scale data infrastructure for storing and processing robot training data at exabyte scale, supporting XDOF's data collection systems and operational capabilities for robotics labs.
• XDOF Datasets Platform XDOF-hosted data platform for exploring and downloading research datasets and evaluations, including the ABC dataset family.

Quantifiable outcome

• Reward model cut error by more than half on held-out human demonstrations, and by more than 80% on real robot rollouts
• +3 more outcomes

Named customers2 records

Segments1 record

Ideal customer profiles1 record

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability7 records

Feature8 records

Partnerships

Five partnerships are on record, tiered core.

• UC BerkeleycoreStrategic or Co-development Partner · 18 June 2026Partnership with UC Berkeley's AI Research lab to release ABC-130K, the world's largest open-source bimanual robot manipulation dataset. XDOF collaborated with partners from UC Berkeley on the dataset of 130,000 robot trajectories plus simulation and evaluation data. The company's roots trace to Berkeley AI research, and the co-founders authored GELLO through their university work on low-cost control systems.
• Carnegie Mellon UniversitycoreStrategic or Co-development PartnerAcademic partner in development of ABC-130K dataset, contributing to the largest open-source teleoperation dataset for robot manipulation research.
• MITcoreStrategic or Co-development PartnerAcademic partner in development of ABC-130K dataset, contributing to the largest open-source teleoperation dataset for robot manipulation research.
• Amazon FAR (Fundamental AI Research)coreStrategic or Co-development PartnerPartnership with Amazon's Fundamental AI Research team for development of ABC-130K dataset, the largest open-source teleoperation dataset for robot manipulation.
• I2RTcoreTechnology or IntegrationHardware partner providing YAM arms used in ABC-130K dataset collection. The bimanual station setups use YAM arms from I2RT, an accessible robot setup that has gained popularity in the robotics research community.

Scale indicators11 records

Recent moves6 records

Expansion highlights5 records

Direct peers

• Skild AI: Skild AI is building foundation models for robotics with backing from top-tier investors. Like Physical Intelligence, it is both a potential customer and a competitor — collecting proprietary robot data while developing general-purpose manipulation policies that compete with what XDOF's customers are building.
• Roboflow: Roboflow provides computer vision dataset management and annotation tooling with a large open-source community. It overlaps with XDOF on the data infrastructure layer for robotics/embodied AI, though Roboflow is more vision-centric while XDOF focuses on manipulation trajectories and reward modeling.
• Physical Intelligence: Physical Intelligence is building foundation models for general-purpose robots and has raised significant capital from top-tier investors. It is both a potential XDOF customer (data infrastructure buyer) and a direct competitor, since it is collecting its own large-scale robot manipulation datasets rather than relying on external providers.
• Covariant (Amazon): Covariant built AI for robotic picking and manipulation, was acquired by Amazon, and now drives Amazon's warehouse robotics. Its founder backgrounds overlap with XDOF's, and it represents the in-house data + model approach that AI labs may replicate — directly competitive on the manipulation AI stack.

Emerging players

• Apptronik: Apptronik develops the Apollo humanoid robot and partners with Google DeepMind on foundation models. It represents a potential XDOF customer for manipulation data but also an in-house data collection competitor in the humanoid robotics space.
• Figure AI: Figure AI builds humanoid robots and is developing its own AI/data infrastructure for general-purpose manipulation. It is a potential customer for XDOF's data services but also a competitor collecting proprietary teleoperation data for its own foundation model efforts.
• 1X Technologies: 1X Technologies is building humanoid robots backed by OpenAI and other investors. It is both a potential XDOF customer and a competitor, collecting its own manipulation datasets while developing foundation models for home and commercial humanoid applications.

Broad incumbents

• NVIDIA (Isaac/GR00T): NVIDIA's Isaac robotics platform and GR00T foundation model for humanoids provide simulation, data, and model infrastructure that competes broadly with XDOF's offering. While much larger and more diversified, NVIDIA's robotics stack threatens to subsume specialist data infrastructure providers by offering an end-to-end platform.
• Scale AI: Scale AI is the largest AI training data provider, offering data labeling and annotation services across modalities. While broader than XDOF (text, image, autonomous driving), Scale is the most analogous business model — managed data services for foundation-model labs — and is increasingly moving into robotics data, directly competing with XDOF.
• Applied Intuition: Applied Intuition provides simulation, data, and tooling infrastructure primarily for autonomous vehicles and increasingly robotics. Founded by ex-Stanford autonomous driving talent, it is broader than XDOF but operates in the adjacent physical-AI tooling space and serves many of the same type of customers (robotics + autonomy labs).

Market position

Strengths5 records

Weaknesses5 records

Competitive moat3 records

Key risks3 records

Key highlights7 records

Customer concentration

• LinkedInlinkedin.com/company/xdofai

Revenue estimate

Valuation estimate

Number of profiles

Profiles4 records

Funding overview

Funding rounds3 records

Investors8 records

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 xdof