Unitree G1-D Standard A U1 Fixed Base Dual-Arm Wheeled Humanoid Robot (Coming Soon)

Unitree G1-D Standard A U1 Fixed Base Dual-Arm Wheeled Humanoid Robot (Coming Soon)

Rather than being marketed solely as a standalone “humanoid,” G1-D is framed as an integrated system that combines a mobile base, a liftable vertical column, and two 7-DoF arms (with optional end-effectors), together with a software workflow that supports data processing, model training, simulation evaluation, and deployment.

In common product listings, the term “fixed-base dual-arm wheeled humanoid robot” usually refers to the platform’s wheeled, ground-operating configuration (as opposed to legged locomotion). In Unitree’s own specifications, wheeled mobility and richer navigation sensing are associated with the G1-D Flagship configuration, which includes a differential-drive base and additional base-mounted sensors.

As an embodied robotics system, G1-D is typically used for tasks such as pick-and-place, bin picking, tool use, multi-camera perception, imitation learning, and policy training for manipulation—especially where repeated real-world trials and scalable dataset acquisition are key requirements.

Design and Features

Mechanical architecture

G1-D’s core mechanical design is built around three main subsystems:

• Mobile base + lift column: Unitree lists two size envelopes depending on whether the column is at minimum or maximum height (approximately 1260 mm to 1680 mm in height).

• Dual-arm manipulation: Both configurations list 7 degrees of freedom per arm (excluding the end effector).

• Waist articulation: Unitree specifies 2 waist DoF, with a wide joint motion space (including Z ±155° and a Y-axis range).

This structure supports a common “mobile manipulation” pattern: the base handles gross positioning, while the lift column and arms handle reach, dexterity, and precise interaction.

Payload and end-effector options

Unitree lists a single-arm maximum load of ~3 kg (noting that payload varies significantly depending on arm extension posture).

End-effectors are presented as selectable options, including:

• Two-finger gripper

• Three-finger dexterous hand (non-tactile)

• Three-finger dexterous hand (tactile)

• Five-finger dexterous hand

These options allow G1-D to be configured for different research goals—from robust industrial grasping to dexterous manipulation and tactile learning.

Technology and Specifications

Configurations: General vs Flagship

Unitree publishes separate parameters for:

• G1-D General (通用版): Approx. 50 kg, 17 DoF (excluding end effectors), and does not list wheeled base speed/drive parameters.

• G1-D Flagship (旗舰版): Approx. 80 kg, 19 DoF (excluding end effectors), and includes a specified wheeled base mobility stack.

Mobility and base sensing (Flagship)

For the Flagship base, Unitree lists:

• Base speed: 1.5 m/s

• Drive: dual-wheel differential drive, supporting 360° in-place rotation

• Base sensors: 1× LiDAR, 2× base depth cameras, plus multiple collision/obstacle recognition sensors (including physical collision sensors and low-obstacle recognition sensors).

This sensor package is characteristic of indoor mobile manipulation platforms that must localize, avoid obstacles, and plan approach trajectories while maintaining arm safety envelopes.

Perception, compute, and connectivity

Unitree lists:

• Compute (base): 8-core high-performance CPU (both versions)

• Perception sensors: 1× head stereo camera and 2× wrist cameras

• Wireless: Wi-Fi 6 and Bluetooth 5.2

• High-compute module: NVIDIA Jetson Orin NX 16GB (100 TOPS) (listed for both)

Power and runtime

Unitree lists:

• Batteries: a quick-release upper-body battery (9Ah) and (for the base) an internal chassis battery (30Ah)

• Runtime: approximately 2 hours (General) vs 6 hours (Flagship)

Development and lifecycle features

The platform is described as supporting:

• OTA updates

• Secondary development (developer enablement)

• A broader workflow that spans data processing → training → simulation evaluation → deployment.

Applications and Use Cases

Embodied AI data collection and imitation learning

G1-D is positioned as a system for building real-world datasets—capturing synchronized streams from head and wrist cameras, base sensors (Flagship), and robot state/action logs for training manipulation policies. The combination of mobile base, dual arms, and configurable end effectors supports collecting diverse interaction data across tables, shelves, bins, and indoor environments.

Robotics research and university labs

Typical users include robotics labs and R&D teams that need:

• A repeatable mobile manipulation platform

• Multi-camera perception for grasping and pose estimation

• A path from “prototype policy” to “robot deployment” using integrated tooling

Service and light industrial scenarios

Unitree’s page highlights broad application domains such as service, living, commerce, and industry, consistent with mobile manipulation applications like item retrieval, shelf scanning, assisted handling, and environment interaction.

Advantages / Benefits

• Mobile manipulation in one platform: Differential-drive navigation (Flagship) combined with dual-arm reach enables tasks that fixed industrial arms cannot perform without conveyors or repositioning fixtures.

• Configurable dexterity: Options from 2-finger grippers to 5-finger hands (including tactile variants) support a wide spectrum of grasping and dexterous manipulation research.

• Rich perception layout: Head stereo plus wrist cameras provide complementary viewpoints, improving grasp success and enabling learning from visual observations.

• Training workflow orientation: The product page emphasizes an end-to-end workflow including simulation-based evaluation and deployment, aligning the platform with embodied AI development cycles.

FAQ Section

What is the Unitree G1-D fixed-base dual-arm wheeled humanoid robot?

The Unitree G1-D is a mobile dual-arm manipulation platform designed for data collection and model training in embodied AI. It combines a wheeled base (Flagship), lift column, dual 7-DoF arms, multiple cameras, and optional dexterous hands.

How does the Unitree G1-D work?

G1-D operates as a mobile manipulator: the base positions the robot in the environment, while the lift column and arms execute manipulation. Its sensors (head stereo camera, wrist cameras, and—on the Flagship—LiDAR and base depth cameras) provide perception inputs for teleoperation, autonomy, or learning-based control.

Why is the Unitree G1-D important?

G1-D targets a major challenge in robotics: collecting high-quality real-world interaction data and rapidly iterating on models. Unitree presents an integrated workflow from data processing to training, simulation evaluation, and deployment, supporting faster experimentation cycles.

What are the benefits of the Unitree G1-D?

Key benefits include dual-arm manipulation, optional dexterous/tactile hands, multi-view perception (head + wrists), and—in the Flagship—1.5 m/s differential-drive mobility with LiDAR and depth cameras for navigation.

Summary

The Unitree G1-D is a dual-arm, mobile manipulation platform optimized for data collection and embodied AI training, offered in General and Flagship configurations. With selectable end effectors (including tactile and five-finger options), multi-camera perception, and an official emphasis on an end-to-end training workflow, it is positioned for research teams and developers building real-world manipulation policies and mobile interaction behaviors.

DESCRIPTION:

Dimensions (Minimum Height): Approx. 1260x500x500mm
Dimensions (Maximum Height): Approx. 1680x500x500mm
Weight with Battery: Approx. 50kg
Total Degrees of Freedom (Excluding End Effector): 17 (Single Arm 7*2 + Waist 2 + Column 1 = 17)
Single Arm Degrees of Freedom (Excluding End Effector): 7
Maximum Arm Payload: Approx. 3kg
Arm Reach (excluding end effector): Approx. 0.45m
Waist Degrees of Freedom: 2
Waist Joint Movement Range: Z±155°, Y -2.5°~+135°
Lifting Stroke: 500mm; Lifting Accuracy: 1mm; Lifting Speed: Max 60mm/s
Maximum operating height: approx. 2m
Equipped with multiple sensing sensors
Cooling system: Localized air cooling
Power supply: Battery-powered or direct power cable connection
Battery runtime: 2 hours
Base computing power: 8-core high-performance CPU
High-performance computing module: NVIDIA Jetson Orin 16G (100 TOPS computing power)
4-microphone array: Supported
5W speaker: Supported
Wi-Fi 6, Bluetooth 5.2: Supported
Smart battery (quick-release): 9000mAh
Charger: 54V 5A
Handheld remote controller: Included
Smart OTA updates: Supported
Secondary development: Supported
Visualization computer: Included

Dex1-1 Gripper Advanced Edition*2
Dimensions: 143x78x67mm
Weight: 550g
Stroke: 90mm
Position Resolution (Tip): 0.1mm
Clamping Force: Adjustable 5~120N
Gripping Load (Conformal): 20kg
Gripping Load (Friction): 5kg
Minimum Full-Stroke Closing Time: 0.24s
Gripper Length: 80mm
Communication Protocol: High-Speed 485

Multi-Sensor Perception System
Head-Mounted High-Definition Binocular Camera*1:
“FOV: H:115° V:80° D:125°; Resolution: 3840×1200”
Wrist-Mounted High-Definition Camera*2:
“FOV: H:130° V:60° D:160°; Resolution: 1920×1080”

Total degrees of freedom: 19 (Single arm 7×2 + Waist 2 + Column 1 + Gripper 1×2 = 19)