Unitree SDK: ROS2, Python, C++ — Where a Developer Should Start

In short: start with Python via unitree_sdk2_python — it's the fastest path to a first working program. For serious development, real-time control and ROS2 integration, move to C++ via unitree_sdk2. The MuJoCo simulator is for debugging without hardware; Isaac Sim is for GPU-accelerated model training and large experiments. Everything lives on the Unitree Robotics GitHub — open code, active community, regular updates.

This article is for developers starting with Unitree (Go2 EDU, G1 EDU, H1, Z1). It explains the SDK architecture and helps you pick the first language/tool.

What's in unitree_sdk2

• unitree_sdk2 (C++) — the basis of real-time control, low-level DDS messages.
• unitree_sdk2_python — Python wrapper for prototyping and training.
• unitree_ros2 — ROS2 packages (Humble and newer) for integration with a standard ROS stack.
• unitree_mujoco — MuJoCo-based simulator, free and fast.
• unitree_rl_gym — reinforcement learning environment on Unitree robots.
• Documentation and examples on GitHub: step-by-step guides, videos, ready scripts.

Which language to start with

Python — for most tasks

If you're starting out — pick Python. unitree_sdk2_python lets you connect to the robot and send a movement command in 10 lines. Fits: first experiments, teaching students, integration with AI/ML libraries (PyTorch, JAX, TensorFlow), fast prototyping.

C++ — for production and real-time

When you need low latency (control at 1000 Hz), integration with existing C++ code, or work with the low-level API — move to C++. This is the standard for industrial applications, real-time motion control, custom low-level behavior.

ROS2 — for robotics-stack integration

If you work in a team using ROS2 or plan to integrate with MoveIt, Nav2, RViz — use unitree_ros2. This is the standard path for university labs and integration into automation systems with other ROS-compatible components.

Simulators: MuJoCo vs Isaac Sim

First steps: what to do in a week

1. Day 1: Install Python 3.10+, clone unitree_sdk2_python from GitHub, run a hello world script.
2. Day 2–3: Install MuJoCo, run unitree_mujoco with a G1 or Go2 model, play with basic commands.
3. Day 4–5: Install ROS2 Humble, clone unitree_ros2, try integration with MoveIt.
4. Day 6: If you have hardware — connect to the real robot via DDS, run the same scripts as in the simulator.
5. Day 7: Explore RL examples (unitree_rl_gym) or AI (OpenVLA on G1) — pick your specialisation.

quadruped

Unitree Go2 EDU

Новое существо воплощённого AI

Baseline start: Go2 EDU + Python SDK. Enough for most education and research tasks.

humanoid

Unitree G1

Гуманоидный AI-аватар

For humanoid development: G1 EDU U1+ with open SDK access.

arm

Unitree Z1

6-осевой робот-манипулятор

Z1 + Python SDK — for manipulation, pick-and-place, imitation learning.

FAQ