Unitree Humanoid Robots

Full-Stack Solution for Embodied

Data Collection and Training

Turn Real-World Task Data into Deployable Robot Capabilities Faster

Watch Video

Contact Sales →

High-Quality Real-World Data Determines Whether Robots Can Enter the Real World

Embodied AI robots must continually learn in real physical environments, developing an understanding of the world through perception, action, and feedback. High-quality, diverse data is the essential foundation for advancing robot capabilities and iteratively improving models.

  • Insufficient High-Quality Demonstration Data

    High-quality interaction data is difficult to obtain in complex physical environments, making it challenging to continuously build training datasets that cover multiple tasks and scenarios.

  • Lack of Embodied Data Collection Workflow Management

    Collection equipment, task workflows, and data management lack support from a unified platform. Fragmented workflows and heavy reliance on manual effort make large-scale, standardized, sustainable data production difficult.

  • Low Efficiency in Converting Data into Models

    Disconnected data processing, model training, and deployment workflows, together with the lack of an efficient training and inference toolchain, hinder rapid data-to-model conversion and continuous iteration of robot capabilities.

Full-Stack Solution
From real-world embodied data collection to model deployment, four core capabilities form a closed-loop data workflow, accelerating capability iteration and real-world application of embodied intelligence robots.
  • High-Precision Teleoperation Interface

    Enables remote interaction between people and robots, providing an operator interface for robot task execution and embodied data collection.

    Learn More

  • Systematic Embodied Data Collection Tools

    An intelligent, one-stop platform covering the entire workflow, including embodied data collection, processing, annotation, review, and data asset management.

    Learn More

  • Comprehensive Model Training and Inference Tools

    Supports distributed model training, custom model development and deployment, and compatibility with mainstream open-source model frameworks.

    Learn More

  • High-Performance Humanoid Robot Platforms

    Multiple types of in-house-developed humanoid robot platforms integrate proprietary motors, reducers, encoders, and a range of robot sensors to meet task requirements across robot form factors.

    Learn More

High-Precision Teleoperation Interface

Through low-latency, high-precision full-body motion interaction, it enables real-time remote control of robots and provides a stable and reliable operational entry point for embodied data collection.

  • Low-Latency Control Response

    Through a low-latency control link, it enables real-time transmission of operation commands and rapid response of robot movements, ensuring synchronized human-robot movements and improving teleoperation smoothness for complex tasks.

  • High-Precision Full-Body Motion Mapping

    It fully captures the operator's upper-limb operations and lower-limb movement states and precisely maps them to the robot's full-body joints, achieving high-fidelity, highly consistent motion reproduction.

  • Synchronized Multimodal Embodied Data Collection

    During teleoperation, it synchronously collects multidimensional information such as first-person-view visual data, pose, environmental perception, and robot state, forming complete demonstration data covering the scene, behavior, and control commands.

  • < 100ms

    System response latency

    Approx. 20ms

    Full-body motion capture latency

  • 60Hz

    Embodied data collection frequency

    ≤ 5cm

    Average position error

    ≤ 6°

    angular error

  • 200Hz

    Motion tracking frequency

    ≥ 98%

    Stepping detection accuracy

Systematic Embodied Data Collection Tools

Through standardized processes and reconfigurable scenarios, it improves the efficiency of scenario-based embodied data collection and reduces the cost of traditional collection methods.

  • Visual Template Management for Efficient Data Collection

    Integrating project management, task assignment, progress tracking, and status analysis, it uses template-based configuration to generate embodied data collection tasks with one click, tracks status in real time throughout the process, enables smoother collaboration, and improves collection efficiency.

  • Flexible Configuration of Diverse Robot Platforms and Data Collection End Effectors

    With “diverse compatibility” as its core design principle, it supports embodied data collection with diverse robot and end-effector configurations and provides strong data standardization capabilities, ensuring end-to-end connectivity from diverse devices to high-quality model training data.

  • Scalable Multi-Robot Collaborative Data Collection

    It provides the technical capability to support multiple robots conducting embodied data collection synchronously. Through a high-concurrency architecture and load-balanced scheduling, it ensures real-time reception and processing of massive data streams, laying the foundation for building high-quality datasets.

  • Stable, Reliable Online Data Collection

    Built on a highly available service architecture, the platform offers strong format compatibility; data can be directly output or converted into various mainstream training formats, improving R&D efficiency.

1

Task design

2

Embodied data collection

3

Quality control

4

Data asset management

5

Training-format delivery

01 / 05

Task design

Establish repeatable task templates and collection standards;

Comprehensive Model Training and Inference Tools

It supports workflows from data processing through model training,simulation evaluation,and one-click deployment. The platform seamlessly integrates multiple mainstream open-source robot model frameworks, enabling developers to efficiently complete the entire R&D and deployment process on a unified platform.

  • A Rich Ecosystem Supporting Mainstream Models

    It builds an open model ecosystem with built-in community datasets, supports training on open-source datasets, and deeply integrates open-source models such as PI, GR00T, and Unitree UnifoLM.

  • Simulation Environments for Model Evaluation

    It includes a high-fidelity, high-precision 3D asset library and rapidly generates comprehensive evaluation plans by building simulation environments, providing support for algorithm validation.

  • Easy to Use, Quick to Deploy

    Ready to use out of the box, it quickly starts model development through “one-click training” and uses integrated simulation tools for reliable evaluation, ultimately enabling seamless migration of algorithms to real robots.

  • Distributed Training for High Performance

    Built on a high-performance distributed training architecture, the platform supports elastic scheduling and parallel acceleration of compute tasks, scales resources dynamically with workload, and achieves 90% GPU utilization.

1

Model architecture selection

2

Training configuration

3

Real-time monitoring

4

Parameter optimization

5

Simulation testing

6

Model export and deployment

01 / 06

Model architecture selection

Select the model architecture and technical approach suited to the task.

High-Performance Humanoid Robot Platforms

  • G1 Series

    A highly agile humanoid robot platform for embodied intelligence research exploration, commercial application validation

  • R1 Series

    A lightweight humanoid robot platform for embodied intelligence algorithm validation, teaching and hands-on training, and commercial

  • H2 Series

    A full-size humanoid robot platform for precision industrial operations and intelligent manufacturing.

  • H2-X Series

    A new generation of full-size humanoid robots, designed for demanding and complex industrial applications.

G1

Full-body motion with a high number of degrees of freedom, dexterous manipulation, multimodal perception, and open development.

Product Parameters

Overall Dimensions

1340×450×200 mm

Overall Weight

Approx. 35kg

Total Degrees of Freedom (Excluding End Effectors)

23~25

Maximum Arm Payload

3~4 kg

Battery Runtime

Approx. 2h

Perception Sensors

Stereo Camera + Chest-Mounted Wide-Angle Camera + 3D LiDAR

G1

G1-A / G1-D

Learn More →

G1-A / G1-D

Multi-form-factor task deployment, operation across an expanded workspace, and stable, continuous execution.

Product Parameters

Overall Dimensions

Approx. 635×450×200 mm ~ 1680×620×620 mm

Overall Weight

Approx. 22~90 kg

Total Degrees of Freedom (Excluding End Effectors)

19~21

Maximum Arm Payload

3 kg

垂直作业空间

0~2 m

Battery Runtime

Approx. 2~6 h

Perception Sensors

Stereo Camera + Chest-Mounted Wide-Angle Camera;

The Flagship Version’s base is equipped with LiDAR, depth cameras, and safety sensors.

G1

G1-A / G1-D

Learn More →

Flexible Configuration, with Data Collection, Training, and Inference Ready Out of the Box
Combine components freely to meet your needs, with hardware and software supplied in one place. Ready-to-use infrastructure lets you focus on your core business.

Configuration Module

  • Multi-Form-Factor Robot Platforms
  • Full-Body Teleoperation Kit
  • High-Performance Private Embodied Data Collection Platform
  • Compute Server
  • Model Training Capabilities
  • 200+

    Task Datasets

    (covering different end effectors and tasks)

  • 4

    Training Algorithms

    (ACT, PI0.5, PI0, and GR00T)

  • 7

    UnifoLM foundation models

  • 10+

    Demos

Application Scenarios
  • Business serivce
  • Housework
  • Education and reserch
  • Industrial application
Start with Real-World Tasks and Build a Closed-Loop Embodied Intelligence Data Workflow
Tell us your company name, contact person and contact details, target robot form factor, target tasks, expected data volume, and delivery timeline. Our solution team will recommend data collection methods, equipment combinations, delivery standards, and an implementation approach.
  • Contact Sales:
    +86 17621781139
  • Contact Team:
    sales_global@unitree.com
  • Contact Address:
    China Digital Valley Building, No. 276 Wulianwang Street, Binjiang District, Hangzhou City, Zhejiang Province, P.R.China