Humanoid robot Unitree G1-D Ultimate D-U9

Unitree G1-D Ultimate D-U9 for Data Collection and Model Training

Unitree G1-D Ultimate D-U9 Humanoid Robot is a humanoid mobile robot on a wheeled differential base, designed for data collection, manipulation algorithm testing, and autonomous navigation. The robot can independently move around the work area without external drive, perform a 360° on-the-spot turn, and move at speeds up to 1.5 m/s. The design features 31 degrees of freedom, two five-finger dexterous hands BrainCo Five-Finger Dexterous Hand Revo 2 Tactile Edition with tactile sensory perception and built-in RGB camera, a retractable column, and a mobile base for autonomous movement. The robot is equipped with an NVIDIA Jetson Orin 16GB computational module with a performance of 100 TOPS, a main HD binocular camera, two wrist cameras, and LiDAR on the chassis. Each arm has 7 degrees of freedom, a working radius of about 0.45 m, and can handle loads up to 3 kg, allowing for complex object grasping and manipulation operations.

Key Features of Unitree G1-D Ultimate D-U9

• Total Degrees of Freedom 31 DoF. The design includes 7 degrees of freedom for each arm, 2 degrees of freedom for the waist, 1 degree of freedom for the vertical lift column, 6 degrees of freedom for each BrainCo five-finger hand, and 2 degrees of freedom for the mobile base chassis.
• Mobile Base for On-Site Work. The two-wheeled differential drive supports a 360° on-the-spot turn and a movement speed of up to 1.5 m/s.
• Retractable Column for Changing Working Height. The platform height varies from 1260 mm to 1680 mm, with a column lift speed of approximately 60 mm/s.
• Base Sensors for Navigation. The chassis features LiDAR, two depth cameras, two physical collision sensors, and two low obstacle sensors.
• BrainCo Revo 2 Tactile Edition Five-Finger Dexterous Hand. Each hand is equipped with a BrainCo Five-Finger Dexterous Hand Revo 2 with 6 degrees of freedom, 11 joints, repeatability accuracy of 0.1°, a height of 160 mm from the tip of the middle finger to the palm base, and tactile sensors for pressure, friction, and force direction sensing. Each hand includes a built-in RGB camera for additional visual data collection.
• Vision System for Data Collection. The platform is equipped with one main HD binocular camera, two HD wrist cameras, and two built-in RGB cameras in the BrainCo hands.
• Computation for Robotic Tasks. The base platform features an 8-core high-performance CPU, and the enhanced performance module includes an NVIDIA Jetson Orin NX 16GB with a performance of 100 TOPS.
• Wireless Communication Support. Unitree G1-D supports Wi-Fi 6 and Bluetooth 5.2.
• Support for Updates and Development. The platform supports Upgraded Intelligent OTA and Secondary Development.
• Working Height up to 2 m. The lift column with a 450 mm stroke allows changing the working area height and performing manipulations at heights up to 2 meters.
• Precise Column Positioning. The column provides positioning accuracy up to 1 mm, which is important for robotic model training and manipulative operations.
• Comprehensive Multi-Sensor System. The platform combines a stereo camera, wrist cameras, RGB cameras in the hands, LiDAR, depth cameras, finger tactile sensors, and collision sensors for real-time data collection and navigation.
• Autonomous Operation from Two Power Systems. The upper module uses a quick-release 9000 mAh battery with up to 2 hours of operation, while the chassis features a built-in 30 Ah battery with up to 6 hours of autonomy.
• User Interaction Tools. The robot supports an array of 4 microphones and a built-in 5W speaker for voice interaction and audio notifications.

Design and Functionality

Unitree G1-D Ultimate D-U9 is a mobile robotic platform with a retractable column, two manipulators with five-finger hands, and a multi-sensor perception system. The robot's dimensions are approximately 1260x525x570 mm in the lower position and 1680x525x570 mm in the upper position, with a net weight including the battery reaching 80 kg.

BrainCo Revo 2 Five-Finger Tactile Hand

Unitree G1-D Ultimate D-U9 comes with two BrainCo Five-Finger Dexterous Hand Revo 2 Tactile Edition five-finger dexterous hands. Each hand has 6 degrees of freedom and 11 joints, allowing for natural human hand movements. The repeatability accuracy is 0.1°, and the hand height is 160 mm.

The Hantic Sensing tactile system provides pressure, friction, and force direction perception on the fingers, which is critically important for performing delicate manipulations, grasping fragile objects, and training tactile interaction models. Each hand's built-in RGB camera provides additional visual control of the work area directly from the finger level.

The manipulation system, along with 7-DoF arms, 2-DoF waist, lift column, 6-DoF hands, and 2-DoF chassis, forms a total of 31 degrees of freedom. Each arm has a working radius of about 0.45 m and a load capacity of up to 3 kg. The waist provides a range of motion Z: ±155°, Y: -2.5° ~ +135°, and the lift column with a stroke of 450 mm, accuracy of 1 mm, and speed up to 60 mm/s allows reaching a working height of approximately 2 m.

Multi-Sensor Perception System

The robot is equipped with a comprehensive multi-sensor system, including a main stereo camera with a resolution of 3840x1200 pixels and viewing angles H:115°, V:80°, D:125°, two HD wrist cameras with a resolution of 1920x1080 pixels and viewing angles H:130°, V:60°, D:160°, two built-in RGB cameras in the BrainCo hands for detailed visual data collection at the finger level, one LiDAR on the chassis, two depth cameras, physical collision sensors, and a low-profile obstacle detection system. Tactile sensors in each finger provide data collection on pressure, friction, and force direction during manipulations. The combination of these sensors ensures three-dimensional mapping of the environment, distance determination to objects, control of the manipulators' work area, and collection of multimodal data for training computer vision models, tactile interaction, and autonomous navigation.

Maneuverability and Robot Movement

The mobility of Unitree G1-D Ultimate D-U9 is ensured by a two-wheeled differential chassis measuring 570x525x197 mm, which has 2 degrees of freedom and supports a 360° on-the-spot turn. The maximum movement speed is up to 1.5 m/s, allowing for quick changes in working positions during research or training scenarios. For navigation, LiDAR, depth cameras, and an obstacle sensor system are used, enabling the implementation of algorithms for autonomous movement, localization, and real-time route planning.

Computational Power and Model Development

Unitree G1-D Ultimate D-U9 is focused on data collection and robotic model training. The computational platform is based on an 8-core high-performance processor and an NVIDIA Jetson Orin 16G module with a performance of up to 100 TOPS, allowing for local processing of large volumes of data from cameras, LiDAR, tactile sensors, and other sensors. The robot supports secondary development, OTA updates, wireless interfaces Wi-Fi 6 and Bluetooth 5.2, and comes with a separate computer for visualization and data processing. Additionally, an array of 4 microphones and a built-in 5W speaker are provided.

ROS 2, Python, C++, DDS middleware, GStreamer for video transmission, as well as distributed model training and algorithm deployment are supported. Available are Gigabit Ethernet, USB, GPIO, and NVIDIA Jetson GPIO.

Power System

Unitree G1-D Ultimate D-U9 has separate power systems for the upper module and the mobile chassis. The upper part can operate from a battery or through direct cable connection and is equipped with a quick-release 9000 mAh battery, providing up to 2 hours of autonomous operation. The chassis uses a built-in 30 Ah battery or power from a charging station, providing up to 6 hours of autonomous movement. The kit includes a main module charger 54V 5A and a chassis charging station 51V 10A. This architecture allows for efficient powering of computational modules, sensors, manipulators, and the movement system during long research sessions.

Application Possibilities

Unitree G1-D Ultimate D-U9 is designed for tasks related to data collection, model training, and algorithm testing in a real environment. The mobile platform with five-finger hands and tactile sensors opens up wide possibilities for exploring fine motor skills, manipulations, and autonomous navigation.

• Collection of tactile and visual data for humanoid robotics
• Training and verification of fine motor skills and object grasping models
• Testing motion control algorithms and five-finger hand
• Researching human-robot interaction (HRI) in a dynamic environment
• Working with open-source datasets and models (PI, GROOT)
• Service, Life, Retail, and Industry scenarios
• Research tasks with a mobile humanoid platform
• Development of solutions for precise manipulations with tactile feedback

Compatibility

Unitree G1-D Ultimate D-U9 supports working with various robotic platforms and configurations. The BrainCo Revo 2 Tactile Edition hand is compatible with tactile data collection protocols and supports integration with manipulation training systems. The platform works with major open-source frameworks of robotic models, open-source datasets, as well as PI and GROOT models. For connection, Wi-Fi 6 and Bluetooth 5.2 are provided.

Package Contents

Unitree G1-D Ultimate D-U9 may include an assembled robotic platform on a wheeled base, two BrainCo Five-Finger Dexterous Hand Revo 2 Tactile Edition hands with RGB cameras, a removable upper part battery 9000 mAh, a built-in chassis battery 30 Ah, a charger 54V 5A, a charging station for the chassis 51V 10A, a handheld control remote, a visualization computer, a computational module based on NVIDIA Jetson Orin 16GB 100 TOPS, and a set of documentation. The exact supply composition and availability of additional modules need to be clarified when ordering.

Technical Specifications

Operating Recommendations

• Plan work sessions considering the operating time: approximately 2 hours for the upper part and approximately 6 hours for the chassis.
• For manipulation tasks, check the arm position, as the maximum load depends on its reach and pose.
• Use Hantic Sensing tactile data for calibrating the grip force for specific objects.
• RGB cameras in the BrainCo hands provide detailed visual control — consider their viewing angle when planning tasks.
• For navigation tests, leave space for the base to turn on the spot 360°.
• Before data collection, set up the task sequence: creation, assignment, collection, annotation, verification, storage, and export.
• For machine learning tasks, check model compatibility with open-source datasets, PI, or GROOT.

Buy Unitree G1-D Ultimate D-U9 Humanoid Robot

Order Unitree G1-D Ultimate D-U9 with price, specifications, and package details clarification before payment. To order the platform, check the availability and required configuration of BrainCo hands.