Unitree Go2

Source: YouTube · Unitree Robotics

Open video on YouTube

Go2 is a four-legged mobile robot designed to operate across indoor and outdoor environments.

Its four-legged architecture gives it an important advantage over wheeled robots: it can maintain mobility across uneven terrain, stairs, slopes, gravel, and other surfaces where conventional wheels can become less effective.

At approximately 15 kg with its battery, Go2 is also small enough to be transported relatively easily.

That makes it fundamentally different from larger professional quadrupeds.

Instead of being designed exclusively around industrial inspection or public-safety applications, Go2 occupies a middle ground between robotics research platform, educational robot, developer kit, and consumer-oriented robotic device.

This broader positioning is one of the reasons the Go series has become closely associated with Unitree's identity.

The most important feature of Go2 may not be its maximum speed or its LiDAR.

It is the price-to-capability ratio.

Quadruped robots have existed for years, but many advanced platforms were developed primarily for universities, military research, industrial inspection, or enterprise customers.

Go2 helped move similar robotic concepts into a much lower price range.

A starting price of $1,600 does not make Go2 a conventional consumer gadget, but it changes who can realistically experiment with quadruped robotics.

A university lab, independent developer, robotics club, technology enthusiast, or small startup can approach the platform without the budget normally associated with advanced mobile robots.

That accessibility is arguably Go2's biggest contribution to the robotics ecosystem.

Go2's mechanical design is centered around twelve articulated joints.

The robot uses aluminum-alloy components and high-strength engineering plastics, creating a relatively lightweight structure while retaining the mechanical strength required for dynamic movement.

Unitree lists a peak joint torque of approximately 45 N·m for the Go2's highest-torque joint configuration.

The robot can perform a wide range of movements, including:

• Walking
• Running
• Turning
• Jumping
• Sitting
• Standing
• Rolling and recovery movements
• Dynamic poses
• Terrain traversal

The company also demonstrates advanced learned gaits, including adaptive recovery and obstacle-climbing behaviors.

However, demonstrations should be interpreted carefully.

A robot's ability to perform a movement in a controlled demonstration does not necessarily mean that the same movement can be performed reliably in every environment.

Go2 is highly capable for its size, but it remains a robotics platform rather than a fully autonomous general-purpose machine.

One of the most important differences between Go2 and simpler robot dogs is its perception system.

Go2 incorporates Unitree's 4D LiDAR L2, which the company describes as providing a 360° × 96° hemispherical field of perception, with a minimum detection distance as low as 0.05 meters.

This allows Go2 to perceive its environment in three-dimensional space while also incorporating motion-related information from the sensing system.

For a mobile robot, perception is fundamental.

A robot dog needs to know not only where it is, but also:

• Where obstacles are
• How far away objects are
• Which surfaces are traversable
• Where it can place its feet
• How the environment changes as it moves

The LiDAR system therefore plays a much more important role than simply acting as an accessory.

It is part of the infrastructure that allows Go2 to move from remote-controlled quadruped toward a more autonomous robot.

Go2 can use its LiDAR system together with the dedicated application to construct a point-cloud map of a defined area.

Users can then specify paths for the robot to follow autonomously.

This transforms Go2 from a robot that simply responds to a controller into a platform capable of performing basic autonomous navigation.

For robotics developers, mapping is especially valuable because it provides a foundation for experimenting with:

• Autonomous navigation
• Localization
• Obstacle avoidance
• Path planning
• Environment mapping
• Human-robot interaction

The exact capabilities available depend on the hardware and software configuration.

Go2 also includes Unitree's ISS 2.0 Intelligent Side-follow System.

The system uses wireless vector positioning and control technology to allow Go2 to follow a person or moving target while combining positioning with obstacle-avoidance strategies. Unitree states that its positioning technology was improved by 50% and that the system supports a control distance of more than 30 meters under favorable conditions.

This is one of the features that makes Go2 feel more like a robotic companion than a conventional remote-controlled machine.

Instead of continuously steering the robot manually, the user can allow the robot to maintain a relationship with a moving target.

That creates possibilities for:

• Personal robotic companions
• Mobile filming
• Outdoor exploration
• Demonstrations
• Human-following experiments
• Robotics research

Unitree increasingly positions Go2 within the broader field of embodied intelligence.

The company says that large-scale AI simulation training has been used to teach Go2 advanced movement patterns, including upside-down walking, adaptive roll-over recovery, and obstacle climbing.

This reflects a broader change in robotics.

Traditional robots often rely heavily on explicitly programmed movement rules.

Modern robots increasingly combine:

Sensors  →  Perception  →  Learned behavior  →  Motion control

The robot learns how to respond to physical situations rather than requiring every possible movement to be manually programmed.

Go2 therefore provides a useful example of how machine learning is increasingly becoming part of the physical control stack.

In addition to LiDAR, Go2 includes an HD wide-angle camera.

The Go2 application provides real-time image transmission and remote monitoring, while certain configurations include 4G and eSIM connectivity for more flexible remote communication.

This makes the camera useful for more than simple recording.

It can support:

• Remote observation
• Teleoperation
• Visual navigation experiments
• Robotics demonstrations
• Mobile inspection
• First-person robotic exploration

The availability of 4G in supported configurations also makes it possible to think of Go2 as a remotely accessible mobile platform rather than a robot that must always remain within local Wi-Fi coverage.

Go2 uses a removable battery system.

Unitree currently lists an 8,000 mAh standard battery, while a higher-capacity battery is available for extended operation. The company's product page gives approximately 2–4 hours of endurance for its long-endurance configuration, while actual runtime depends heavily on operating conditions.

The standard battery is rated at 8,000 mAh and 236.8 Wh, according to Unitree's battery documentation.

As with most mobile robots, advertised endurance should not be interpreted as a fixed operating time.

Running, climbing, high-performance movement, wireless communication, computing, and repeated dynamic actions can consume significantly more energy than slow walking or standby operation.

For extended robotic work, spare batteries remain an important consideration.

Go2 integrates a dedicated onboard compute stack alongside its high-performance joint motor controllers, enabling real-time sensor fusion, LiDAR mapping, locomotion planning, wireless telemetry, and OTA firmware updates across the Wi-Fi 6, Bluetooth, and optional 4G / eSIM connectivity options. Optional high-computing-power expansion modules allow developers to add more advanced perception, SLAM, and AI model inference workloads on-device.

Unitree currently lists several Go2 configurations, including Air, Pro, X, and EDU.

The differences are important because the Go2 name refers to a product family rather than one completely standardized configuration.

Because Unitree continues to offer different hardware combinations, buyers should compare the exact configuration rather than choosing purely based on the Go2 name.

One of Go2's strengths is the growing accessory ecosystem.

Unitree currently lists optional hardware including:

• High-computing-power modules
• Small servo arm
• Intel RealSense D435i depth camera
• Additional 3D navigation LiDAR
• Charging station
• Additional batteries
• Remote controller
• Tracking module
• Foot-end components

This modular approach changes how the robot can be used.

A basic Go2 can function primarily as a mobile quadruped, while additional sensors and computing hardware can turn it into a platform for more advanced robotics experiments.

Go2 is not only intended to be controlled through a conventional remote.

Unitree also provides graphical programming capabilities, allowing users to build behaviors using visual programming concepts rather than writing every component from scratch.

Depending on the configuration, the platform can also support deeper development and customization.

This makes Go2 interesting across several levels of technical experience:

The exact level of access depends on the Go2 configuration and development environment.

The easiest way to understand Go2 is not to compare it only by specifications.

Think of it as occupying three categories simultaneously:

This combination is what makes Go2 more interesting than a simple remote-controlled robot.

Go2 also represents something larger than the robot itself.

The next generation of AI is increasingly moving beyond screens.

A language model can generate text.

A vision model can understand images.

An embodied AI system must eventually perceive the physical world and act within it.

A robot such as Go2 provides a physical body for exploring that problem.

It has:

Sensors  →  Perception
AI models  →  Decision making
Control algorithms  →  Movement
Motors  →  Physical action

This makes Go2 an interesting platform for experimenting with embodied intelligence at a scale that is much more accessible than many larger humanoid or industrial robots.