Unitree G1 Humanoid Robot

The G1 is a small-form-factor humanoid robot built around a human-like two-leg and two-arm configuration.

At approximately 1.32 meters tall and 35 kilograms with its battery, it is considerably smaller and lighter than many full-size humanoid robots. That gives the platform a different practical profile: it is easier to transport, store, demonstrate, and deploy in research environments where a full-size industrial humanoid may be unnecessarily large.

Its relatively compact body is also part of Unitree's broader strategy of making advanced robotic hardware more accessible to developers and researchers.

The G1 was officially introduced on May 13, 2024, and has since become one of Unitree's best-known humanoid platforms.

The most interesting aspect of G1 is not any single specification.

Its significance comes from the combination of humanoid form, dynamic movement, relatively compact hardware, and a comparatively low entry price.

Humanoid robots have traditionally been expensive research platforms. G1 helped push the category toward a different model: a humanoid that can be purchased by universities, robotics labs, developers, and technology enthusiasts without necessarily requiring an industrial-scale budget.

That makes G1 particularly relevant to the emerging ecosystem around embodied AI.

Instead of treating the robot purely as an autonomous machine, developers can use the physical platform to experiment with perception, locomotion, manipulation, imitation learning, reinforcement learning, and human-robot interaction.

Movement is one of G1's defining characteristics.

The robot uses high-performance joint motors and a wide range of joint motion to produce movements that go well beyond simple walking. Unitree lists 23 degrees of freedom for the standard G1, while the G1 EDU can be configured with up to 43 degrees of freedom.

The standard configuration includes:

• Six degrees of freedom per leg
• Five degrees of freedom per arm
• One degree of freedom at the waist

The mechanical design also uses industrial-grade crossed roller bearings and dual encoders, while the joints use low-inertia, high-speed permanent-magnet synchronous motors.

This combination gives G1 a relatively large movement envelope for its size.

Its capabilities are therefore not limited to conventional humanoid walking. The platform has been demonstrated performing dynamic movements, balancing motions, getting up from the ground, and other physically demanding actions.

The important distinction is that these demonstrations show the robot's physical capability, not necessarily that every movement is available as a reliable autonomous function for everyday users.

G1 is designed around the idea that modern humanoid robots should improve through software as well as hardware.

Unitree identifies imitation learning and reinforcement learning as important technologies behind the robot's movement capabilities.

Imitation learning allows a robot to learn behaviors from demonstrations, while reinforcement learning can be used to optimize movement policies through simulation and training.

This approach is particularly important for humanoids because programming every possible movement manually would be extremely difficult.

Walking, recovering balance, manipulating objects, and coordinating multiple joints are all problems that can benefit from learning-based control.

As a result, the G1 should be viewed less like a conventional appliance and more like a physical AI development platform.

G1 combines several types of sensors to understand its surroundings.

The official specification lists:

• A depth camera
• A 3D LiDAR
• A four-microphone array
• Other onboard sensing associated with its control system

The depth camera and 3D LiDAR provide complementary information about the robot's environment, helping support perception and navigation tasks. The microphone array provides an audio input channel for voice and interaction-related applications.

This sensor combination is particularly useful for robotics research because the robot needs more than joint-position information to operate in a real environment.

A humanoid must understand both where its body is and what is around it.

Humanoid robots become significantly more useful when they can manipulate objects rather than simply walk.

This is one reason the G1 platform can be configured with dexterous hands.

The G1 EDU can support a three-finger force-controlled dexterous hand with additional degrees of freedom. Unitree describes the system as using force-position hybrid control to provide more precise interaction with objects.

This opens the door to experiments involving:

• Object grasping
• Tool manipulation
• Pick-and-place tasks
• Human-robot interaction
• Imitation learning
• Fine motor control

However, it is important to distinguish hardware capability from general-purpose autonomy. Having a dexterous hand does not mean that G1 can independently perform arbitrary household or industrial tasks without additional software, training, and integration.

The standard G1 includes an 8-core high-performance CPU and supports Wi-Fi 6 and Bluetooth 5.2.

The G1 EDU expands the platform further by supporting high-computing-power modules, including options based on NVIDIA Orin-class hardware, according to Unitree's current specifications.

This distinction is important for developers.

The standard G1 is positioned primarily as a ready-to-use robot, while the EDU version is intended for users who want to integrate their own software, computing hardware, and robotics research workflows.

Unitree lists approximately two hours of battery life for the G1.

In practice, runtime depends on what the robot is doing. Walking, dynamic movement, manipulation, computing workload, and environmental conditions can all affect power consumption.

The robot uses a removable 9,000 mAh smart battery, making battery replacement more practical for extended demonstrations or development sessions.

At approximately 35 kg, however, G1 is still a substantial piece of equipment. It may be compact compared with a full-size humanoid, but it is not a lightweight consumer gadget.

G1 makes the most sense for users who want to experiment with humanoid robotics rather than simply own a humanoid robot.

It is particularly relevant to:

For buyers interested primarily in custom software, manipulation research, or hardware integration, the G1 EDU is likely the more appropriate platform.

For users primarily interested in experiencing a capable humanoid robot without extensive development work, the standard G1 provides a simpler entry point.

The importance of G1 extends beyond its individual specifications.

Unitree originally became widely known for quadruped robots such as the Go series. With G1, the company moved deeper into humanoid robotics while maintaining the same general philosophy: combine advanced robotic hardware with aggressive cost reduction and relatively compact designs.

That makes G1 an interesting marker in the evolution of humanoid robotics.

The question is no longer simply whether humanoid robots can walk, balance, or perform impressive movements. The larger question is whether these machines can become useful, trainable, scalable physical AI platforms.

G1 is one of the platforms being used to explore that question.