UNITREE G1-D

End-to-End Platform
for Humanoid Robot

Data & Training: A complete ecosystem accelerating humanoid robotic R&D processes.

Core Components

High-Performance Humanoid Robot

High-Performance Humanoid Robot

Multiple proprietary humanoid embodiments, featuring in-house actuators, gearboxes, encoders, and sensors.

Streamlined Data Acquisition Tools

Streamlined Data Acquisition Tools

A unified platform for the complete data pipeline: acquisition, processing, labeling, review, and data asset management.

Comprehensive Model Training & Inference Tools

Comprehensive Model Training & Inference Tools

Enables distributed training, custom model development, and seamless deployment, with full support for all major open-source frameworks.

G1-D Platform

Higher-DOF Robot Platform

Total DOF (Excl. End-Effector)

19

Arm Degrees of Freedom

7×2

Waist Degrees of Freedom

2

Column Degrees of Freedom

1

Base Degrees of Freedom

2

Expanded Operational Workspace

Lower-Latency Control Response

Streamlined Data Acquisition Tools

Streamline data collection and reduce costs with flexible, standardized workflows — moving beyond traditional manual processes.

Visual Template Management for Efficient Collection

Integrating project management, task allocation, progress tracking, and status analysis into a unified platform. With templated configurations, data collection tasks can be generated with one click.

Flexible Configurations with Diverse Platforms & End Effectors

The system supports data collection across various robot platforms and end-effector configurations. Its robust data standardization capability enables end-to-end processing to high-quality model training data.

High-Concurrency Architecture for Scalable Operations

Equipped with the technical capability to support hundreds of robots performing synchronous data collection. Through high-concurrency architecture, the platform ensures real-time processing of massive data streams.

24/7 Online Collection, Stable and Reliable

Built on a highly available service architecture, the platform enables reliable 24/7 data collection. Collected data can be directly converted into mainstream training formats.

Data Acquisition Pipeline
1

Creating Collection Tasks

2

Task Editing & Assignment

3

Acquisition & Annotation

4

Upload & Review

5

Data Storage

6

Data Export

Step 1
Step 2
Step 3
Step 4
Step 5
Step 6

Comprehensive Model Training & Inference Tools

It supports an end-to-end workflow from data processing to one-click model deployment. The platform seamlessly integrates a variety of mainstream open-source robotic model frameworks.

Rich Ecosystem with Mainstream Model Support

An open model ecosystem is established, featuring built-in community datasets and support for open-source dataset training. It offers deep integration with leading open-source models such as PI and GROOT.

Simulation Environment for Model Evaluation

Includes a high-fidelity, high-precision 3D asset library. By constructing realistic simulation environments, it rapidly generates comprehensive evaluation schemes to support algorithm validation.

User-Friendly Design for Rapid Deployment

Fully deployable out-of-the-box, minimizing setup time. Launch model development instantly via 'one-click training,' and achieve seamless migration from algorithms to real-world machines.

Distributed Training for High Efficiency

Built on a high-performance distributed training architecture, the platform enables elastic scheduling of computing tasks and parallel acceleration. Achieving up to 90% GPU utilization.

Training and Inference Pipeline
1

Model Architecture Selection

2

Training Configuration

3

Real-Time Monitoring

4

Parameter Editing

5

Simulation Testing

6

Export & Deployment

Step 1
Step 2
Step 3
Step 4
Step 5
Step 6

UnifoLM-WMA-0

A World-Model-Action (WMA) Framework

UnifoLM-WMA-0 is Unitree's open-source world-model-action architecture designed for general-purpose robot learning. Its core is a world-model capable of understanding physical interactions.

Decision-Making Mode

Predicts future physical interactions to generate actions, reducing decision errors.

Simulation Mode

Interactive simulator generating high-fidelity feedback based on robot motion inputs.

World models are one of the central approaches of the 2025–26 robot-learning wave — NVIDIA's DreamDojo is in the same category and officially supports the Unitree G1. We cover the concept, pipeline and data requirements in our DreamDojo World Model Guide (Turkish).

Application Scenarios

Service

Service

Life

Life

Retail

Retail

Industry

Industry

Live Installation
& Testing

Watch the installation and field tests of the Unitree G1-D development platform. Setup processes and initial robot responses conducted by our authorized technical team.

Field Setup & Calibration
UnifoLM-WMA-0 Software Integration
Autonomous Motion Tests

Product Parameters

ParameterG1-D (Standard)G1-D (Flagship)
Dimensions (Min. Column)1260x500x500mm1260x525x570mm
Dimensions (Max. Column)1680x500x500mm1680x525x570mm
Total WeightApprox. 50kgApprox. 80kg
Total DOF1719
Single Arm DOF77
Max. Arm PayloadApprox. 3kgApprox. 3kg
Max. Speed/1.5m/s
Battery LifeApprox. 2 hoursApprox. 6 hours
Basic Computing Power8-core CPU8-core CPU
Yüksek İşlem Gücü ModülüOrin NX 100TOPSOrin NX 100TOPS

* Parameters may vary according to different scenarios.

The Future of
Robotic Development

Request a Demo