Unitree G1 - ROBOTS: Your Guide to the World of Robotics

The Unitree G1 is a compact humanoid robot designed for AI-powered mobility and manipulation. It stands out for its agility, force-controlled dexterous hands, and the ability to perform tasks like backflips and fine object handling.
Unitree trained the G1 to perform competitive Tai Chi routines—demonstrating not just agility but smooth, human-like motion control.
Founded in 2016 by Xingxing Wang, Unitree Robotics emerged from academic research focused on low-cost, high-performance legged locomotion. While pursuing his master’s degree, Wang independently developed the XDog—a quadruped robot using low-cost external rotor brushless motors and a full-degree-of-freedom actuation system. His work included custom motor drive board design, control architecture development, and simulation-based validation of robot gait dynamics. This approach, emphasizing affordability without sacrificing capability, has since become a widely adopted development strategy in quadruped robotics. Unitree’s early commercial platforms, including Laikago and Aliengo, brought legged robots to a broader audience, with applications in research, inspection, and mobility R&D.
In 2023, Unitree entered the humanoid robotics field with the launch of the H1, described as China’s first full-size general-purpose humanoid robot capable of running. The H1 demonstrated a high power-to-weight ratio relative to its peers and signaled the company's broader ambition to move beyond quadrupeds.
On May 13, 2024, Unitree followed up with the G1, a smaller humanoid platform introduced as an "AI avatar." Measuring approximately 127 centimeters tall and weighing 35 kilograms, the G1 features articulated limbs with 23 to 43 degrees of freedom, depending on the configuration. The robot also impressed with its price: US $16,000—significantly lower than that of previous humanoids with comparable capabilities.
The G1 humanoid is built using Unitree's custom actuator technology and leverages high-torque joint modules similar to those found in its quadrupeds. Early demonstration footage showed the G1 performing dynamic locomotion tasks, including backflips, as well as fine object manipulation, highlighting the robot's mechanical flexibility and software-defined control.
The G1 represents a convergence of high-performance mechatronics and embodied AI in a relatively compact form factor. The robot is designed for extensibility and modular control—featuring multi-axis compliance, inverse-kinematics-based motion planning, and real-time balance correction. Its relatively low mass and self-contained actuators may make it attractive for academic labs and robotics developers exploring general-purpose humanoid robots in research, teleoperation, and AI‑driven robotics tasks.
Though still in an early stage, the G1’s launch positions Unitree among a growing field of companies—such as Boston Dynamics, Tesla, Agility Robotics, and Figure AI—attempting to commercialize bipedal AI-powered robots for logistics, service, and human-machine interaction.
A full-size, general-purpose humanoid robot capable of dynamic bipedal locomotion, including running.
Payload capacity for arms of about 2 kg (base model). EDU model increases payload to 3 kg and expands degrees of freedom up to 29-43 DoF, including additional wrist axes and a 7‑DoF force‑controlled dexterous hand.
Livox Mid-360 3D lidar plus Intel RealSense D435i depth camera. Microphone array. Dual joint encoders (for position/torque feedback) and IMU to support balance and motion stabilization.
Low inertia high-speed internal rotor permanent magnet synchronous motor. 90 Nm maximum knee joint. (EDU model extends to 120 Nm.)
Lightweight, high-strength aluminum and carbon fiber parts. Industrial-grade crossed-roller bearings.
Base model: 8-core high-performance CPU. EDU model: NVIDIA Jetson Orin compute modules delivering up to 100–157 TOPS of AI processing. Wi-Fi 6, Bluetooth 5.2.
Software stack with force‑position hybrid control for dexterous manipulation using three‑finger hand. Learning methods include imitation learning and reinforcement learning with Unitree's UnifoLM (Unified Large Model) framework. Compatible with ROS 2, and simulation tools like Isaac Sim or MuJoCo.
9,000 mAh lithium battery, 2 hours of operational runtime