Unitree Robotics has unveiled an innovative wheeled robot

Unitree Robotics has announced its first wheeled humanoid robot, the G1-D, designed for advanced data collection and training of artificial intelligence systems. This robot acts as the central component of a comprehensive platform that combines powerful hardware and scalable software for skilled data collection, AI training and real-world solution deployment.

The robot comes in two versions: standard and flagship. Both versions range in height from 1,260 to 1,680 mm and weigh up to 80 kg. The standard G1-D is designed for stationary applications, while the flagship version features a mobile platform with differential drive that allows speeds of up to 1.5 meters per second. The mobile base includes lidar, depth cameras and collision sensors.

The G1-D is equipped with 17 degrees of freedom in the basic configuration and 19 in the extended configuration, not including end effector devices. Each arm contains seven degrees of freedom and can support loads up to 3 kg. The belt joint has a range of motion in the Z-axis up to ±155° and in the Y-axis from -2.5° to +135°. The maximum height of the working platform is 2 meters.

It has a high-resolution binocular camera on its head, complemented by HD cameras on its wrists. It allows the use of different types of gripping devices: two-finger hands, three-finger hands with or without tactile sensors, and full five-finger hands. The central component of the computing system is the Nvidia Jetson Orin NX module – with the ability to achieve up to 100 TOPS (trillions of operations per second). The flagship version can run autonomously for up to six hours.

The G1-D is designed to perform a variety of tasks: automating manual operations in industry and services, working in warehouses, collecting and annotating data for AI training, and interacting with stores and retail outlets thanks to its built-in vision and mobility. Flexible arms combined with a mobile platform allow the robot to manipulate objects, conduct inspections or act as a data collection agent in challenging environments.

Besides the hardware solutions, Unitree has developed a complete software stack to efficiently manage the robot’s training and operational processes. It includes tools for data collection, task management, annotation and real-time tracking systems. The platform supports distributed learning, integrates with open source frameworks such as GROOT and PI, and includes a modeling environment for model evaluation. All of this enables seamless export and deployment of AI-trained models.