Choose a humanoid research platform based on the access your experiment actually needs: the ability to run your own control software, access to sensor data, or a workable route from simulation to a physical robot. Before ordering, confirm the rights and features available for the exact model variant. For example, the manufacturer does not support further development on the standard Unitree G1, and recommends the EDU Edition for this purpose. For longer-term lab projects, data flows, documentation for the specific delivery and clear service terms should also factor into your decision.
Which research teams is it for?
Teams working on motion and learning algorithms
If you develop your own control systems or learning methods, you’ll need permission to run your own software, as well as access to control and state data. In its R1 developer documentation, Unitree lists interfaces including SDK, Robot-State and ROS2. Your team should confirm before purchase that the required access is available for the variant on offer.
Labs working on perception and data capture
For experiments involving image, audio or teleoperation data, sensor interfaces are only part of the picture. You’ll also need to consider export options, access rights and how the data is processed. If people can be identified in the data, you’ll need to plan appropriate safeguards; the German Data Protection Conference highlights data minimisation in particular for scientific research.
Project teams planning long-term lab use
Ongoing experiments call for planned training, repair arrangements and points of contact. Terra Robotics describes procurement, commissioning, training and technical support for Unitree. The project agreement should set out which services and timescales apply to the specific model.
Which requirements matter most?
If your lab is carrying out its own development, start by establishing what access the experiment requires. You can then assess data processing, simulation and operation over the project’s lifetime for the exact model variant.
| Requirement |
What to check before buying |
Who it matters most to |
| Your own software and control |
Get written confirmation of the SDK and usage rights for the exact variant, as well as the required access to control and state data. |
If we want to run our own motion or learning algorithms, pre-programmed routines won’t be enough. For around 33% of buyers, this development access is the main priority. |
| Sensors and data |
Clarify access to raw data and its formats, along with whether image, audio and teleoperation data is processed locally or externally. |
For our perception experiments, we need to capture data without losing control over who can access it and how it is processed. Around 27% of buyers put this data access first. |
| Simulation and teleoperation |
Check the simulation model and version compatibility; confirm the teleoperation function and any equipment required for the specific model. |
We want to test procedures virtually first, then carry them out on the robot in a controlled way. For around 20% of buyers, this transition is the main criterion. |
| Service throughout the project |
Set out training, repair arrangements, points of contact and spare-parts terms in the quote or service agreement. |
For recurring experiments, we need to be able to plan for downtime and repairs. Around 20% of buyers prioritise reliable service terms. |
Common selection mistakes
-
Assuming a standard model is an EDU variant
Unitree states that the standard G1 does not support further development, and recommends the EDU Edition if you need to customise it. So, a shared product family name doesn’t tell you what software development is possible.
Ask for the full model name, SDK scope, licence and required access rights to be included in the quote.
Choosing an unsuitable variant can mean extra integration work or having to switch platforms.
-
Confusing sensor access with clear data-processing arrangements
A streaming interface alone doesn’t tell you who can access image, audio or teleoperation data, or where it is processed. For research involving personal data, the German Data Protection Conference emphasises the need for appropriate safeguards and data minimisation.
Before running experiments involving identifiable people, check data flows, storage locations, access roles and deletion options for the specific configuration with your data protection officer.
-
Treating simulation as proof of safety in real-world use
A virtual trial is no substitute for checking the physical robot and its surroundings. PAL Robotics recommends testing behaviour in simulation first for its VR teleoperation, and notes that this feature may not be available as standard.
Confirm that the feature is available, and plan the move to real-world use with your own checks of the control system, network and safety.
Testing and approval on the physical system will take additional project time.
-
Mistaking a general support offer for a project-specific commitment
Terra Robotics describes replacement, repair and support services. That doesn’t establish agreed response times or spare-parts lead times for a specific robot model.
Agree repair location, points of contact, software support and required timescales in writing before ordering.
Unclear repair arrangements can delay recurring experiments.
Relevant brands to consider
Unitree
Unitree offers humanoid models and EDU variants for development and research. What matters is which features are included with the specific variant on offer.
- For the R1, Unitree lists open joint and sensor interfaces, as well as simulation support. The R1 developer documentation includes SDK, DDS, video streaming and ROS2, among other features.
- Unitree does not support further development on the standard G1. Development rights and interfaces therefore need to be confirmed for the specific EDU or research variant.
PAL Robotics
PAL Robotics describes TALOS as a configurable, ROS-based humanoid platform for research into movement and whole-body control.
- TALOS’s ROS-based approach is relevant to labs that build their development environment around ROS. PAL Robotics also documents simulation and teleoperation for its platforms.
- Before ordering TALOS in Germany, check the configuration on offer, available features, service terms and documentation for the specific delivery.
Related products at Joybuy
Related products coming soon.
Frequently asked questions
Is a standard Unitree G1 suitable for running your own research software?
Not for further development: Unitree explicitly states that the standard G1 does not support it, and recommends the EDU Edition for customisation. Ask for written confirmation of SDK rights and the interfaces you need for your chosen EDU configuration.
What should you check before experiments involving cameras, microphones or teleoperation?
If people could be identified in the data, you’ll need to clarify how it will be processed and what safeguards are appropriate for the specific experiment. The German Data Protection Conference particularly emphasises data minimisation in scientific research. You should also check the platform’s data flows, access rights, storage locations and deletion options.
Which documents matter when buying for a German lab?
For relevant electrical or radio features, check the documentation for the specific delivery. The German Federal Network Agency lists CE marking, manufacturer details, a German-language user manual and a declaration of conformity among its market surveillance checks.
Is a general support promise enough for a long-term research project?
A general promise is no substitute for a project-specific agreement. Terra Robotics describes training and technical support, as well as replacement and repair services for Unitree. Set out the response times, repair arrangements and spare-parts terms required for your model in the quote or contract.
Does a research platform automatically include VR teleoperation?
No. PAL Robotics describes its documented VR teleoperation as a premium feature that may not be available on a particular robot. Check the model configuration, licence and required equipment before planning to use this feature in an experiment.