Motion control
Standing, walking, turning, getting up and recovering from pushes.
What matters
T1T2H2 EDUJoint-level access, simulation, and a protective rig for low-level work.
Research
Humanoid platforms, from first policy to deployment.
Match your research task to the right platform, edition and toolchain. We confirm software versions and interfaces before you order, then handle sourcing and delivery.
Start from the question.
Four directions, and what each asks of a platform. Most projects touch more than one.
Perception and interaction
Recognising objects and people, visual localisation and voice interaction.
What matters
T1T2H2 PlusCamera access, microphones and enough on-board computing.
Manipulation
Target localisation, hand-eye coordination, grasping and two-handed tasks.
What matters
T2H2 Plus7-DoF arms, a gripper or dexterous hand, and wrist cameras.
Simulation and learning
Reinforcement and imitation learning in simulation, then transfer to hardware.
What matters
T1T2H2 PlusRobot models, simulators, external GPUs and a sim-to-real plan.
Choose by the task.
Four platforms, from developer-led motion work to the full embodied AI workflow. Figures follow the manufacturers’ published tables.
Booster T1
Developer platform for motion control, perception and robot football.
- Height · weight
- 1.18 m · approx. 30 kg
- Degrees of freedom
- 23 (base configuration)
- Computing
- Basic: Jetson AGX Orin 32GB, 200 TOPS
- Arms
- 4 DoF each; 7-DoF version available
Booster T2
Flagship platform for manipulation and integration research.
- Height · weight
- approx. 1.4 m · approx. 42–43 kg
- Degrees of freedom
- 31 · 7 per arm · 3 waist
- Computing
- 375 TOPS (INT8) or 2,070 TFLOPS (FP4, sparse)
- Hands
- Standard, gripper or dexterous hand
Unitree H2 EDU
Full-size humanoid; only the EDU edition opens secondary development.
- Height · weight
- 1820 mm · approx. 70 kg with battery
- Degrees of freedom
- 31 (body)
- Computing
- User development unit; Thor module optional
- Hands
- Dexterous hands optional
Unitree H2 Plus
Built for the full embodied AI workflow, from teleoperation to deployment.
- Height · weight
- 1820 mm · approx. 70 kg with battery
- Degrees of freedom
- 31 body; 75 with two optional Sharpa Wave hands
- Computing
- Jetson T5000 · 2,070 TFLOPS (FP4, sparse) · 128 GB
- Hands
- Sharpa Wave hands and wrist cameras optional
Computing figures use different units and precisions, so they are not directly comparable. Optional parts depend on your order; we confirm current availability for each configuration.
From data to the real robot.
A typical embodied AI workflow, with the tools documented for each platform. Each link needs checking on your exact versions.
-
01/04
Collect
Record demonstrations by teleoperation: camera images, joint states and actions.
xr_teleoperate · Isaac TeleOp -
02/04
Simulate
Test motions and agents in a virtual environment before using hardware.
Booster Studio · MuJoCo · Isaac Sim -
03/04
Train
Train policies with reinforcement or imitation learning, on your own GPU workstation.
unitree_rl_gym · Isaac Lab · LeRobot -
04/04
Deploy
Run the policy on a protected robot, then measure the sim-to-real gap.
SDK · ROS 2 · Jetson
We record firmware, SDK and ROS 2 versions with your order, and check each repository’s supported robots against your configuration.
Teleoperation and data.
An operator guides the robot; the recording becomes training data. Every step from headset to policy needs testing on your setup.
Teleoperation
Control the arms from an XR headset
Unitree’s xr_teleoperate supports Apple Vision Pro, PICO 4 Ultra Enterprise and Meta Quest 3, including H2 with 7-DoF arms. H2 Plus isn’t on that list; its page lists Isaac TeleOp.
H2H2 Plus
Data
From recordings to a trained policy
unitree_lerobot converts and replays recordings in LeRobot v3 format for training. Before you rely on it, we suggest a short test: record, convert, train and replay on your exact configuration.
H2H2 PlusImages in this section are illustrations, not photographs of a specific setup.
What a demo shows.
Demonstrations show what a platform can be used for, not what ships in the box. Here is what each one still needs.
-
Two-handed work on H2 Plus. Hands and wrist cameras are optional, and some functions are still in development.
-
Picking up and moving boxes. A real task needs your data, a trained policy and testing on site.
-
Tabletop manipulation in a research scene. It shows a setting for experiments, not housework ability.
-
Dance and movement routines on H2. Each motion needs an action pack or policy matched to the model.
-
T2 end effectors. Grasping results still depend on perception, control software and testing with your objects.
Images show manufacturer demonstrations. Autonomous and teleoperated results should always be reported separately.
What ships ready, and what your lab builds.
Platforms ship with interfaces and tools. Results come from your team’s work, on versions we record together.
Ready today
Built in
Documented for each platform.
- Walking, balance and preset motionsT1T2H2
- Python SDK: Python 3.10+, firmware v1.7+T1T2
- unitree_sdk2, ROS 2 and DDS interfacesH2 EDUH2 Plus
- Simulation and data replay toolsT1T2
Configure
Depends on edition
Edition, optional parts and software access.
- Secondary development: H2 EDU only, not standard H2H2
- Computing: T2 Education or Professional editionT2
- Wrist cameras on gripper and dexterous-hand versionsT2
- Hands and wrist cameras are optionalH2 Plus
- T1 with 7-DoF arms: no football, teach or dance skillsT1
Develop
Project work
Built, trained and validated by your team.
- Locomotion and whole-body control policiesT1T2H2 EDU
- Grasping, sorting or folding tasksT2H2 Plus
- Teleoperation datasets for your taskH2H2 Plus
- Model training and on-board deploymentT2H2 Plus
Computing by edition
Board-level figures from the manufacturers’ tables. Units differ and cannot be compared directly.
| Processor | AI computing | Memory | |
|---|---|---|---|
| T1 Basic | Jetson AGX Orin 32GB | 200 TOPS | 32 GB |
| T1 Standard | Express-i7 (motion) + Jetson AGX Orin 32GB (perception) | 200 TOPS on the perception board | 8 GB · 32 GB, per board |
| T2 Education | High-performance ARM | 375 TOPS (INT8) | 16 GB |
| T2 Professional | Jetson Thor T5000 | 2,070 TFLOPS (FP4, sparse) | 128 GB |
| H2 Plus | Jetson T5000 | 2,070 TFLOPS (FP4, sparse) | 128 GB unified |
Training workstations, safety rigs and XR headsets are not part of the robot. We list what your project needs alongside the quote.
Tell us about your research.
Share the task, your team and the systems you already use. We’ll check which platform, edition and interfaces fit, and flag anything that needs development.