Unitree G1-D Flagship Edition C Wheeled Humanoid Robot

Get Things Done

The Unitree G1-D is a compact humanoid robot featuring dual articulated robotic arms, a height-adjustable column and an integrated sensor head. Flagship Edition C (U8) comes equipped with dual Dex3-1 grippers, providing an excellent platform for robotics research, AI development and data collection. It arrives in style with a rechargeable LiDAR mobile platform with auto collision and obstacle sensing, and moving upwards, offers upgradeable and downgradable end effector I/O as well. Backed by Unitree's comprehensive software ecosystem and support for major open-source frameworks, the G1-D is ready for everything from data acquisition and model training to deployment.

Higher-DOF Platform

With the robot itself with 17 degrees of freedom, the G1-D delivers the flexibility needed for complex robotic movement and manipulation. Its articulated upper body includes 7 degrees of freedom per arm, 2 waist degrees of freedom, and 1 column degree of freedom, enabling natural, adaptable motion across a wide range of tasks. The added mobile base grants 2 more DOF and the pair of included Dex3-1 grippers grant a total of 14 thanks to their excellent manouverability.

Streamlined Data Acquisition

A unified software platform manages the complete data pipeline, covering acquisition, processing, annotation, review, and data asset management. Visual template management brings project management, task assignment, progress tracking, and status analysis into a single interface, while one-click task generation and real-time workflow monitoring help improve collection efficiency.

Oversee Work with Ease

Supporting multiple robot configurations, the platform standardises data from diverse devices into high-quality training datasets. Its high-concurrency architecture enables hundreds of robots to collect data simultaneously, while a highly available service architecture supports reliable 24/7 operation. Data can be exported directly or converted into mainstream machine learning training formats, helping accelerate development.

Model Training & Inference

Comprehensive model training and inference tools support distributed training, custom model development, and deployment. Full compatibility with major open-source machine learning frameworks gives developers the flexibility to build, train, and deploy models within a single ecosystem.

UnifoLM-WMA-0 Framework

The G1-D supports UnifoLM-WMA-0, Unitree's open-source World-Model-Action (WMA) framework for general-purpose robot learning. At its core is a world model that understands the physical interactions between robots and their environments.

In Decision-Making Mode, the framework predicts future physical interactions from the current environment and task objectives, helping the policy module generate more accurate actions while reducing decision errors. In Simulation Mode, it produces high-fidelity environmental feedback and synthetic training data to support model training and policy optimisation, accelerating the learning process.

• Compact humanoid robot platform featuring dual articulated robotic arms, a height-adjustable column, and an integrated sensor head.
• Features a set of Unitree Dex3-1 7DOF grippers for advanced handling.
• Features a rechargable 2DOF LiDAR mobile platform with obstacle and collision sensing.
• 19 degrees of freedom (excluding end effectors), including 7 DOF per arm, 2 waist DOF, 1 column DOF and 2 base DOF.
• Supports upgradeable end-effectors for different robotic applications.
• Equipped with upgradeable end effectors to support different manipulation and research applications.
• Developed for robotics research, AI development, robot learning, and real-world data collection.
• Combines in-house actuators, gearboxes, encoders, and sensors into a complete humanoid hardware platform.
• Supports Unitree’s open-source UnifoLM-WMA-0 World-Model-Action framework for general-purpose robot learning across multiple robotic embodiments.
• Uses a world model to understand physical interactions between robots and their environments, supporting more capable decision-making and simulation workflows.
• Decision-making mode predicts future physical interactions to help generate more accurate robotic actions.
• Simulation mode produces high-fidelity environmental feedback and synthetic data to support model training and policy development.
• Unified software platform manages the complete data pipeline, including acquisition, processing, annotation, review, and data asset management.
• Visual template management combines project management, task assignment, progress tracking, and workflow analysis in one platform.
• One-click task generation and real-time workflow monitoring simplify large-scale data collection workflows.
• Supports multiple robot configurations while standardising collected data into high-quality training datasets.
• High-concurrency architecture supports simultaneous data collection from hundreds of robots.
• Highly available service architecture enables reliable 24/7 data collection.
• Data can be exported directly or converted into mainstream machine learning training formats.
• Supports distributed training, custom model development, inference, and deployment.
• Compatible with major open-source machine learning frameworks.