Humanoid robot Unitree G1-Comp

Humanoid Robot Unitree G1-Comp Platform for AI Research and Development

Unitree G1-Comp is an advanced humanoid robot specifically designed for participation in international RoboCup competitions, conducting scientific research, and education in the field of mechatronics. The model features a compact foldable design, equipped with advanced spatial vision sensors, and offers a flexible environment for testing machine learning technologies. The platform has 25 degrees of freedom (DoF), two 5-axis arms, 1 degree of freedom in the waist, a multi-sensor perception system, and a built-in computational module NVIDIA Jetson Orin with a performance of 100 TOPS. The robot is suitable for universities, laboratories, R&D teams, and companies working with robotics, computer vision, artificial intelligence, bipedal walking algorithms, and process automation.

Main Advantages of Unitree G1-Comp

• Platform for RoboCup and competitions. Unitree G1-Comp is designed for football robotics tasks, motion control, field orientation, and working with visual data. It is a robot for RoboCup, where stable mechanics, precise sensors, and access to SDK are important.
• Flexible configuration of degrees of freedom. The humanoid robot has 25 active degrees of freedom (DoF) with the possibility of extending kinematics for specific research tasks. The base model distributes actuators anatomically: 6 DoF per leg, 5 DoF per arm, 2 DoF for the head, and 1 DoF for the waist with the option to add 2 more joints.
• High torque in the knee joint. The maximum torque of the knee joint is 120 Nm. This is important for walking, squats, practicing motor algorithms, and scenarios with sharp load redistribution.
• Sensors for navigation and computer vision. The robot is equipped with an Intel RealSense D455 depth camera and an industrial 3D LiDAR scanner for circular space reading. The sensors form a dense point cloud of the environment, providing a base for SLAM localization, dynamic obstacle avoidance, and complex room geometry analysis.
• AI development readiness. The NVIDIA Jetson Orin / Orin NX module provides up to 100 TOPS of computing power for artificial intelligence. This is useful for YOLO11, autonomous behavior, sensor data processing, and working with models on the robot.
• Multimodal interaction UnifoLM. The platform supports working with a proprietary large language model of artificial intelligence for recognizing the operator's natural language. Integrated TTS speech synthesis and ASR automatic audio stream recognition services allow programming live voice dialogue scenarios.
• Two-level control via API. Interfaces for high-level and low-level operation control are available for coding, providing direct access to status services and communication via a fast DDS interface. Engineers can flexibly switch between full control of each individual actuator and sending general movement commands.
• Compatibility with ROS and simulators. The robot is fully integrated into the global ROS / ROS2 ecosystem for quick connection of ready-made navigation packages and third-party sensor drivers. Virtual environments Isaac Gym and Mujoco are supported for preliminary digital testing and safe training of reinforcement models.
• Quick control via mobile app. Proprietary APP software allows for quick basic humanoid configuration and real-time system telemetry tracking. Through a smartphone or tablet screen, you can monitor battery charge status, activate audio-light services, and check communication status.
• Quick-release power Smart Battery. The onboard 13-cell smart battery provides approximately 2 hours of continuous autonomous platform operation. The design with a quick-release system allows for instant replacement of a discharged unit, minimizing pauses during laboratory experiments or competitions.
• Autonomous operation up to 2 hours. The quick-release smart battery with 9000 mAh provides approximately 2 hours of operation. This is convenient for training, educational sessions, laboratory tests, and short demonstration scenarios.
• Ease of transportation. The robot's body has a unique folding feature, reducing its height dimensions by almost half. This allows the device to be transported in a regular car trunk without using special cases.
• Extended official warranty. Basic technical support from the manufacturer is provided for 18 months. Depending on the requirements of a specific league or competition regulations, this period can be officially extended.

Design and Motion Control

Unitree G1-Comp has dimensions of 1320×450×200 mm in a standing position and 690×450×300 mm when folded. The weight with the battery is approximately 35 kg, making the robot suitable for laboratory transportation, educational classes, test sites, and competition venues.

The mechanics are based on 25 degrees of freedom depending on the configuration. One leg has 6 degrees of freedom, one arm 5, the head 2, the waist 1 with an option for 2. The length of the shin and thigh is 0.6 m, the arm span is approximately 0.45 m, and the maximum load on the arm is about 3 kg depending on the arm's position.

The robot supports significant motion angles: waist Z ±155°, X ±45°, Y ±30°; knee 0–165°; thigh P ±154°, R -30°~+170°, Y ±158°; wrist P ±92.5°, Y ±92.5°. Cables are routed in joint cavities, joints have dual encoders, and local air cooling helps during long tests.

Reinforcement Learning and unitree_rl_gym Framework

The open library unitree_rl_gym is used to transfer skills from virtual models to real hardware, offering developers the following functionality:

1. Efficient simulation environment: Full integration with Isaac Gym and Mujoco tools for accelerated physical calculation of humanoid movements.
2. Flexible training configuration: Setting training parameters through simple configuration files with adjustment of the number of parallel environments, random seeds, iteration limits.
3. Full development cycle: Support and automation of the process of transferring motor policies from initial Sim2Sim modeling to final code deployment on a real robot Sim2Real.

Sensors and AI Computing in the RoboCup SDK Sports Environment

Effective orientation on the field requires fast processing of spatial data and clear coordination of actions. The competition software and hardware complex is based on a specialized RoboCup SDK, which provides the following developer tools:

• Visual Recognition API: Features a built-in YOLO11 neural network of the latest generation for real-time video stream processing. Used for quick identification of the ball, stadium lines, goals, and recognition of opponent robots.
• Spatial Positioning API: Provides an interface for precise determination of the humanoid's coordinates on the field. Algorithms combine monocular geometric positioning and binocular depth determination through the D455 sensor.
• Motion Control API: Receives ready coordinates from vision systems and translates control commands directly to the actuators. Ensures stable running, one-on-one maneuvering, push resistance, and precise ball strikes.

Application Areas

• RoboCup and robotic competitions. Unitree G1-Comp is designed for football scenarios, working with RoboCup SDK, visual recognition, and motion control on the field.
• Universities and laboratories. The robot is suitable for research groups studying walking, balance, sensory perception, autonomous behavior, and human interaction.
• Robotics education. This is a robot for training students and teams working with ROS, simulators, APIs, sensors, and robotic algorithms.
• Development of artificial intelligence algorithms. The platform provides computational resources for recognition models, motion planning, data processing, and decision-making.
• Computer vision testing. The D455 camera, 3D LiDAR, and YOLO11 provide a base for object detection, localization, scene analysis, and RoboCup tasks.
• Research on humanoid robot movement. Degrees of freedom, motion ranges, 120 Nm in the knee joint, and dual encoders are needed for walking, balance, and maneuvering tests.
• Demonstration projects. The humanoid robot can work at exhibitions, educational events, and presentations where a real robotic platform with control and sensors is needed.
• R&D departments of companies. Unitree G1-Comp is suitable for testing robotic hypotheses, prototypes, autonomy scenarios, and applied algorithms in development teams.

What's Included in the Unitree G1-Comp Package

The package is designed for launching the robot, charging, and basic manual control. The supply composition can be specified before ordering depending on the configuration.

1. Unitree G1-Comp robot.
2. Quick-release smart battery.
3. Charger.
4. Manual control remote.
5. Documentation and available setup materials, if provided with the supply.

Technical Specifications of Unitree G1-Comp

Buy Unitree G1-Comp

Want to order Unitree G1-Comp for your engineering team or laboratory? During the order, find out the current cost and agree on individual delivery terms. Our technical specialists will answer all questions regarding the SDK architecture, AI module specifications, and help select a configuration for your project tasks.