---
title: "Unitree Go2 Quadruped Robot"
space: "Factory Docs"
url: "https://factory.uga.edu/docs/equipment/unitree-go2"
updated: "2026-11-19"
---

## Overview

The Unitree Go2 is a compact 12-DOF quadruped platform for mobile robotics, autonomy, perception, and embodied-AI research. Our lab has both **Go2 Pro** and **Go2 EDU** units.

The distinction matters for development: the Pro is useful for operation, demonstrations, and higher-level behaviors, while the EDU is the research configuration with supported secondary development and low-level SDK access. Do not assume software written for the EDU can be deployed unchanged to a Pro.

## Pro vs EDU


| Parameter                 | Go2 Pro                                 | Go2 EDU                           |
| ------------------------- | ---------------------------------------: | ---------------------------------: |
| Standing size             | 70 x 31 x 40 cm                         | 70 x 31 x 40 cm                   |
| Weight with battery       | \~15 kg                                 | \~15 kg                           |
| Joint motors              | 12                                      | 12                                |
| Nominal speed             | 0-3.5 m/s                               | 0-3.7 m/s                         |
| Payload                   | \~8 kg; \~10 kg limit                   | \~8 kg; \~12 kg limit             |
| Maximum climbing slope    | \~40 deg                                | \~40 deg                          |
| 4D LiDAR                  | Yes                                     | Yes                               |
| Wide-angle camera         | Yes                                     | Yes                               |
| Foot-end force sensors    | Not standard                            | Yes                               |
| Secondary development     | Limited / not the EDU interface set     | Yes                               |
| High-compute module       | Not the primary developer configuration | Optional, configuration dependent |
| Typical battery           | 8,000 mAh                               | 15,000 mAh long-endurance         |
| Typical published runtime | \~1-2 h                                 | \~2-4 h                           |


> Unitree sells multiple Go2 configurations and revises firmware and options over time. Treat the physical robot, purchase configuration, and installed firmware as authoritative for a specific unit.

## Mechanical Structure

The Go2 has four legs with three actuated joints per leg:

```text
              Body
       /       |       \
    Hip       Hip      ...
     |         |
   Thigh     Thigh
     |         |
   Calf      Calf

4 legs x 3 joints = 12 DOF
```

The EDU specification lists approximately 45 N.m maximum torque for the largest joint motor. Published joint ranges are approximately:

- Body/hip: -48 to +48 deg
- Thigh: -200 to +90 deg
- Calf: -156 to -48 deg

## Perception

The Go2 sensor system includes a wide-angle camera and Unitree 4D LiDAR. EDU units add foot-end force sensing and may be fitted with additional compute or perception hardware.

Typical research pipeline:

```text
LiDAR / Camera / Joint State / Foot Force
                  |
                  v
            State Estimation
                  |
          +-------+-------+
          |               |
          v               v
       Mapping        Perception
          |               |
          +-------+-------+
                  |
                  v
            Task Planning
                  |
                  v
          Locomotion Command
```

## Development Architecture

For Go2 EDU development, Unitree provides `unitree_sdk2` and DDS-based communication. Official developer documentation includes C++ and Python interfaces, ROS 2 services, high-level motion APIs, low-level control examples, LiDAR services, SLAM/navigation interfaces, and robot-state services.

A useful mental model is:

```text
Research Application
        |
        | DDS / unitree_sdk2 / ROS 2
        v
Unitree Application Interfaces
        |
        v
Motion + State Services
        |
        v
Real-Time Robot Controller
        |
        v
 Motors / IMU / Sensors
```

Keep experimental software above the real-time motor-control layer unless low-level control is specifically required.

## High-Level vs Low-Level Control

### High-level control

Use high-level motion APIs when the experiment needs behaviors such as:

- stand / sit
- walking velocity
- body pose
- trajectory execution
- autonomous navigation
- mapping

This is normally the preferred interface because Unitree's controller retains responsibility for balance and locomotion.

### Low-level control

The EDU exposes interfaces suitable for research involving direct joint-level commands and custom locomotion policies.

Low-level development is appropriate for:

- reinforcement-learning locomotion policies
- custom gait controllers
- whole-body control research
- actuator/joint-control experiments

It also removes much of the protection provided by higher-level locomotion software. Test low-level controllers with physical clearance, an operator ready to stop the robot, conservative gains, and an appropriate support/test setup.

## Networking

Go2 supports Wi-Fi 6 and Bluetooth. Development commonly involves a Linux workstation or onboard compute module communicating with robot services over Ethernet/Wi-Fi networking and DDS.

When debugging SDK connectivity, separate the problem into:

1. physical/network link,
2. interface and subnet configuration,
3. DDS discovery,
4. SDK/API compatibility,
5. robot operating mode,
6. application logic.

If basic DDS discovery is broken, changing locomotion code will not fix the problem.

## ROS 2 and DDS

Unitree's current SDK ecosystem uses DDS as the underlying messaging system and provides ROS 2 integration.

Conceptually:

```text
ROS 2 Node       Native SDK Application
     |                    |
     +-------- DDS -------+
               |
               v
        Go2 Robot Services
```

This makes the EDU suitable for integrating perception pipelines, autonomous navigation, external sensors, and experimental control systems without treating ROS as the robot's real-time motor controller.

## Lab Research Use

Our Go2 EDU has already been used as the mobility platform for the **SCOUT** autonomous UHF RFID inventory-monitoring system. That system combines Go2 mapping and mobility with an external Linux compute module and UHF RFID hardware to autonomously patrol an environment and localize RFID tags.

This makes the existing Go2 platform particularly relevant for:

- autonomous inventory monitoring
- UHF RFID localization
- manufacturing logistics
- SLAM and navigation
- mobile perception
- legged locomotion
- sensor fusion
- reinforcement learning
- payload and sensor integration
- human-centric manufacturing research

## Payload Integration

The body can carry research hardware such as:

- embedded compute modules
- cameras
- RFID readers
- antennas
- environmental sensors
- custom electronics

When adding a payload, consider both **mass and mass distribution**. A payload below the nominal mass limit can still destabilize the robot if its center of mass is high, offset laterally, or loosely mounted.

## Battery and Charging

Go2 batteries are removable. EDU configurations commonly use the 15,000 mAh long-endurance battery and fast charger, while Pro configurations commonly use the 8,000 mAh battery.

Runtime depends strongly on gait, payload, terrain, compute load, and battery condition. Do not use published runtime as an experiment-duration guarantee.

## Safety

The Go2 is a dynamically balancing legged robot, not a safety-rated collaborative robot.

Important hazards include:

- sudden motion
- falls
- foot impacts
- leg-joint pinch points
- payload detachment
- loss of balance during experimental control
- high-current batteries
- unexpected behavior from custom software

For new control software:

1. Clear people and fragile equipment from the test area.
2. Start with conservative velocity and joint limits.
3. Keep the remote/emergency control available.
4. Verify the commanded mode before transmitting controls.
5. Avoid simultaneously running competing motion-control applications.
6. Test payload retention before dynamic motions.

## Troubleshooting

### SDK application cannot see the robot

Check:

- correct network interface
- robot and development computer addressing
- multicast/DDS traffic
- firewall configuration
- SDK version
- robot firmware

### Robot receives state but does not move

Check:

- requested robot mode
- motion service ownership
- safety/stop state
- whether another application is publishing commands
- whether the API is supported on the specific Pro/EDU configuration

### Robot behaves unexpectedly under custom control

Immediately stop the experiment and check:

- command units
- joint indexing
- coordinate conventions
- command rate
- gains
- stale commands
- competing DDS publishers
- state-estimation assumptions

Never debug unstable low-level locomotion by repeatedly allowing the physical robot to fall.

## References

- Unitree Robotics, *Go2 SDK Development Guide*.
- Unitree Robotics, *Go2 Product/Specification Comparison*.
- `unitree_sdk2` developer examples and interface documentation.
- Internal SCOUT research documentation for the lab's Go2 EDU implementation.