From Grade R to PhD on one robot family.

Write down what learners must be able to do at the end of the term: describe a sensor, write a loop, build a map, train a model. Then choose the lowest tier that delivers it. The navigator on this page does exactly that.

Base models (Go2 Air and Pro, R1 Air and Basic, G1 Basic and Pro) run Unitree’s app behaviours only, and Unitree offers no upgrade from a base unit to an EDU unit. If coding is in your plan within two years, buy a Go2-X, an EDU tier or an R1-A Smart from day one.

Unitree publishes simulation twins for MuJoCo and NVIDIA Isaac Sim, and the same code runs on the virtual and the real robot. A whole class can code at once in simulation while one robot runs the best programs. It costs nothing and protects the hardware.

One teacher or lecturer owns the robot’s timetable, charging log and firmware updates. A second adult is trained on the remote’s stop and damping controls. Unitree robots are research-grade machines for trained adult operators; learners program, adults operate.

• A marked robot zone of at least 3 × 3 m on a dry, non-slip floor (2 m clearance all round for the H2).
• A dedicated Wi-Fi network or VLAN that isn’t connected to the school’s administration systems.
• A lockable, ventilated cabinet for charging and storing batteries.
• Ubuntu laptops for coding, and for universities an NVIDIA RTX workstation for reinforcement learning.

Order a spare battery with every robot so one pack charges while the other teaches. Stocked models typically arrive 4 to 6 weeks after payment; newer models can take longer, and the quotation confirms the lead time. Order a full term before you need the robot. MCM issues formal 30-day quotations and supports tender paperwork.

1. Week 1: unbox, safety briefing and the e-stop drill.
2. Weeks 2–3: simulate first in Isaac Sim or MuJoCo.
3. Weeks 4–6: first code on hardware, high-level Python, then ROS 2 nodes.
4. Weeks 7–10: perception projects, mapping, vision and person-following.
5. Exam block (universities): joint-level control, reinforcement-learning policies sim-to-real, and the write-up.

1. Term 0, see it move. Book a live demo in George or Midrand, or a virtual one. Name one champion teacher, run this navigator with the HOD and the bursar, and confirm where the money comes from. Cost: an afternoon.
2. Term 1, pilot the platform. One robot, one class, one term: a STEM Classroom Kit (R 133,498) or an R1 Humanoid Starter Lab (R 190,750), ex VAT. Simulation first, then a term-end demonstration for parents and the SGB.
3. Year 1, build the lab. Add stations, spare batteries, a charging cabinet, a gantry for humanoids and a sign-out log, and put robotics on the timetable.
4. Year 2, showcase and compete. Enter a competition, host a district robotics day, partner with a university, and use the results to attract sponsors.

• High-level motion commands: stand, walk, turn and gestures in a few lines of Python or C++.
• Low-level joint control: torque, position and velocity for every motor.
• Raw sensor streams: LiDAR, depth camera, cameras and IMU.
• On-board compute: Jetson modules from 40 to 100 TOPS run your models on the robot.
• ROS 2 on native DDS, plus simulation twins in Isaac Sim and MuJoCo.

All free on github.com/unitreerobotics:

• unitree_sdk2 and unitree_sdk2_python: the core SDK in C++ and Python.
• unitree_ros2: ROS 2 integration for navigation, mapping and visualisation.
• unitree_mujoco: a simulator that runs the same code against a virtual robot.
• unitree_rl_gym and unitree_rl_lab: reinforcement-learning pipelines in NVIDIA Isaac Gym and Isaac Lab, ready to deploy to the robot.
• xr_teleoperate: control humanoid arms and hands from Apple Vision Pro, PICO or Meta Quest, and record demonstrations.
• UnifoLM-WMA-0 and UnifoLM-VLA-0: open robot foundation models released in September 2025 and January 2026.

Blocks through the app in primary school, Python with a simplified safe API in Grade 7 to 9, full Python and ROS 2 from Grade 10 and TVET, C++, ROS 2 and PyTorch at university. The point of one ecosystem is that every step builds on the last.

• Everyday coding: a laptop running Ubuntu Linux connected to the robot’s network.
• Reinforcement learning: a workstation with an NVIDIA RTX graphics card.
• On the robot: the Jetson module on EDU tiers runs inference while it moves.

A class of 30 needs 30 simulation seats, not 30 robots.

ROS 2 Nav2 and SLAM tools for navigation, OpenCV and YOLO-family detectors for vision, NVIDIA Isaac Sim and Isaac Lab for simulation and training, and Hugging Face LeRobot for imitation learning. Universities can teach industry-standard tools on affordable hardware.

Unitree improves its robots through over-the-air updates, including security fixes. Keep firmware current through MCM Robotics, record the version in your robot log, and test an update in simulation-only mode before a class depends on it.

The commands below follow the README of Unitree’s official Python SDK. Configure the robot’s network first, using Unitree’s developer quick-start guide, and keep the robot in a clear zone with the e-stop ready.

# Ubuntu laptop on the robot’s network
sudo apt install python3-pip
git clone https://github.com/unitreerobotics/unitree_sdk2_python.git
cd unitree_sdk2_python
pip3 install -e .

# 1. Check communication (two terminals)
python3 ./example/helloworld/publisher.py
python3 ./example/helloworld/subscriber.py

# 2. Read the robot’s state (use your network interface name)
python3 ./example/high_level/read_highstate.py enp2s0

# 3. First movement: stand up and lie down
python3 ./example/high_level/sportmode_test.py enp2s0

If pip can’t find CycloneDDS, build CycloneDDS 0.10.x first, as the README explains. Run the same logic in the MuJoCo twin before it touches hardware.

Competition builds and the R1 EDU Professional E to H configurations: confirm secondary development on your quotation.

Bundle prices are the sum of the parts in our live catalogue.

Prices read live from our catalogue on 23 October 2026. Other configurations are on each robot’s page.

• Airfreight, customs and VAT, itemised on the MCM quotation.
• Spare batteries and chargers.
• A dedicated router or VLAN, and a lockable charging cabinet.
• Ubuntu laptops; an RTX workstation for universities.
• Teacher release time for training.
• Spares after the warranty period, for example a Go2 joint motor at R 6,562.50.

A STEM Classroom Kit at R 133,498 ex VAT, used for five years by 150 learners a year, costs about R 178 per learner per year. Use the plan calculator above with your own numbers; shared hubs push the figure lower still.

• Provincial Mathematics, Science and Technology (MST) Conditional Grant, which already supports Coding and Robotics, enrichment, teacher development and laboratory resources.
• School governing body budgets.
• Section 18A donations from companies and foundations.
• B-BBEE skills development and socio-economic development spend, and corporate social investment.

• NRF National Equipment Programme, with co-investment from the institution.
• University equipment committees and faculty research budgets.
• SETA discretionary grants for skills programmes.
• Industry partnerships in mining, automotive, agriculture and manufacturing.

1. Configure online or ask for a departmental recommendation.
2. Receive a formal PDF quotation in Rand, valid 30 days, with banking details.
3. Pay by EFT; MCM handles airfreight, customs and SARS clearance.
4. Delivery anywhere in South Africa, typically 4 to 6 weeks for stocked models, with orientation material.

Tender documentation and institutional paperwork are supported.

• A trained adult operates the robot; learners program and observe.
• Run the e-stop and damping drill before every session.
• Mark the robot zone and keep young learners behind a line.
• Use the gantry stand during humanoid development.
• Indoors, dry floors, 0 to 40 °C; these robots aren’t waterproof.
• Never leave a powered robot unattended.

Charge lithium packs on a non-flammable surface, never unattended overnight, and store them in a ventilated cabinet. Stop using a swollen or damaged pack and contact MCM. Ask MCM for the dangerous-goods guidance if a robot travels to competitions.

In September 2025 researchers disclosed a Bluetooth flaw, known as UniPwn, affecting Go2, B2, G1 and H1 firmware; further G1 issues were reported in 2026. Unitree acknowledged the findings. Practical controls for schools and universities:

• Keep firmware current through MCM.
• Switch Bluetooth off once Wi-Fi is configured.
• Run robots on an isolated Wi-Fi network or VLAN without internet by default.
• Change default credentials and never connect robots to administration networks.
• Review and disable cloud features you don’t need.

Robots carry cameras and microphones. Get consent from parents or students, avoid storing identifiable images or voices, keep project data on local machines, and don’t enable cloud recognition features with learners without written approval. University data-collection projects should go through ethics review.

Walking, balance recovery, preset routines and conversation are reliable today. Picking up arbitrary objects is still an active research problem, and the acrobatics in launch videos are trained routines. Saying this up front makes the robot exceed expectations instead of disappointing them.

Backed by Unitree and handled locally by MCM Robotics.

First-line support, repairs and genuine parts run from the George head office (+27 44 878 2917), with sales and operations in Midrand (+27 12 030 3451) and a Cape Town branch opening in November 2026. Wear from normal teaching use is repairable in George.

Orientation material covering safe handling, batteries and the e-stop drill, pointers into Unitree’s official documentation, SDKs and simulators, and the gantry stand on R1 EDU and G1 EDU tiers. MCM is a call away for setup questions.

Book a departmental demo, live or on video, on the contact page. Tell MCM your course outcomes and budget, and you’ll leave with a tier recommendation and a formal quotation in Rand.

• 2025: shipped 5,511 humanoids, the most of any company, and launched the low-cost R1 and the full-size H2.
• September 2025: open-sourced the UnifoLM-WMA-0 world-model-action model; security researchers disclosed the UniPwn Bluetooth flaw.
• 2025: launched a graded Go2 training course for students, with training kits and debugging stations.
• January 2026: open-sourced the UnifoLM-VLA-0 vision-language-action model.
• First half of 2026: number one in quadrupeds with 37% of global shipments (Counterpoint), and about 5,900 humanoids shipped (SAG).
• March to July 2026: STAR Market listing approved, with 2025 revenue up more than 300%.
• August 2026: listed on the Shanghai STAR Market.

• February 2024: Coding and Robotics CAPS for Grades R to 9 approved by Umalusi.
• 2025: Foundation Phase rollout begins.
• 2026: available grade by grade in primary schools where there is capacity.
• March 2026: provincial specialists trained at the CHPC in Cape Town.
• July 2026: Limpopo teams at the Robotics for Good Youth Challenge in Geneva.
• October 2026: MCM Robotics shows live Unitree robots at GovTech 2026 in Durban.

Unitree iterates quickly. Buy the tier your course needs this year, rely on over-the-air updates for new capability, and plan a five-year teaching life. Newer models extend the ladder rather than replacing the skills learners have built.

Yes, when it runs the way it is designed to: a trained adult operator, a marked zone, the e-stop drill and close supervision. Young learners interact by planning, predicting and observing, not by handling the robot.

One robot per class works well with a rotation model: groups code in simulation while one group runs its program on the robot. A district hub on a timetable can serve many schools a year.

No. Robots run on a local network. Firmware updates can be done through MCM, and keeping robots off the open internet is good security practice.

No. Unitree doesn’t offer that upgrade. If coding might be in your plans, buy a programmable tier from the start.

Start in simulation and use the gantry stand for humanoid development to prevent most damage. Normal teaching wear is repairable in George, and genuine parts are priced in Rand. Misuse falls outside the warranty, so the adult-operator rule matters.

Yes. Unitree’s simulators are free. Many schools start a term in simulation and buy, or borrow from a hub, once learners have code worth running.

R1 models carry built-in voice AI; supported languages depend on the software package. Ask MCM to confirm specific languages in writing on your quotation.

Yes. Unitree runs a graded Go2 training course covering maintenance, motion control, navigation and secondary development; OpenMind and RoboStore offer a modular university curriculum for the G1; and online courses teach lecturers ROS 2, simulation and Sim2Real on Unitree robots. Combine them with the six stages and the CAPS crosswalk on this page.

The G1-Comp and G1 EDU Boxing builds are made for competition and boxing events, and MCM’s specification sheet doesn’t list secondary development for them. For a club that wants to code, choose a Go2-X, a Go2 EDU or an R1 EDU, and add a G1 EDU when the club reaches university-level work.

Other good robots exist. Unitree’s advantage for education is the combination of research-grade robots at education prices, one of the largest open-source communities, and one ecosystem from Grade 4 to PhD, supported locally by its official distributor.

Yes. MCM issues formal 30-day quotations, supports tender documentation and invoices in Rand from a South African entity, so procurement never deals with foreign exchange or imports.

Software development kit: the code libraries that let learners program the robot directly. Unitree’s SDK2 works in C++ and Python. Base models don’t expose it; EDU tiers do.

Unitree’s term for programming a robot beyond its built-in app behaviours. EDU tiers have full access, the Go2-X partial access, base models none.

The education and research configurations of a Unitree robot, with on-board compute, developer access and technical support. A base unit can’t be upgraded to EDU later.

Robot Operating System 2: the open-source framework universities and industry use to connect sensors, planners and controllers. Unitree robots support it.

The number of joints a robot can move independently. More DoF means more dexterity, and more to control.

Trillions of operations per second: a measure of on-board AI compute. 40 TOPS runs vision models; 100 TOPS runs larger models in real time.

A laser sensor that measures distances all around the robot and builds a 3D point cloud. Every Go2 carries one.

A camera that measures how far away each pixel is, so the robot can see shape as well as colour.

Simultaneous localisation and mapping: the robot builds a map of a space while working out where it is in that map.

A virtual copy of the robot and its world, in tools such as MuJoCo and NVIDIA Isaac Sim. Learners test code here before the real robot moves.

Moving a behaviour learned in simulation onto real hardware, and closing the gap between the two.

A robot learns by trial and error against a reward, usually millions of attempts in simulation.

A robot learns a task from recorded human demonstrations, often collected by teleoperation.

A person controls the robot remotely, for example through a VR headset, to perform a task or record demonstrations.

AI that acts in the physical world through a body, rather than only producing text or images.

Whatever sits at the end of a robot arm: a gripper, a three-finger hand or a five-finger hand.

Touch sensors in a robot hand that measure contact and pressure, needed for delicate grasping.

The software inside the robot itself. Keep it current through MCM: updates fix bugs and security issues.

A frame that holds a humanoid by a harness while new code is tested, so a fall can’t damage it.

A documented check, on delivery, that the robot, its configuration and its software work as quoted.