Humanoid Robot Buyer's Guide 2026: Compare Models, Costs & Use Cases
How to buy or lease a humanoid robot in 2026: whether you need one, the models you can actually buy (Unitree G1, H1 and H2, Booster T1 and K1, AGIBOT X2) with live US prices, what Figure, Tesla Optimus and Atlas offer instead, real costs and your first 90 days.
An honest guide to the humanoid robot market in 2026: who is actually selling, what works, what does not, and whether you should buy one at all.
The State of Humanoid Robots in 2026
The humanoid market in 2026 has two halves. On one side are robots you can order today, mostly from Chinese makers such as Unitree, Booster Robotics and AGIBOT, sold to labs, schools, event companies and pilot teams. Our humanoid price guide lists every one we sell, with live prices. On the other side are the best-known Western humanoids (Tesla Optimus, Figure, Boston Dynamics Atlas, Agility Digit), which are in internal use, partner deployments or enterprise programs rather than on open sale.
What works: bipedal walking on flat surfaces is reliable across the platforms you can buy. Basic pick-and-place of known objects in a structured setup works on platforms with capable arms. Teleoperation, a human operator controlling the humanoid remotely, works and is the main way teams collect training data for autonomous behaviour.
What does not work yet: autonomous manipulation of novel objects, dependable recovery from falls, operation in genuinely unstructured environments, and multi-step tasks without human intervention. The gap between teleoperated demonstrations and autonomous execution remains large. If you are buying a humanoid in 2026, you are buying a research platform, a data-collection tool or a demonstrator, not an autonomous worker.
Are You Ready for a Humanoid? Five Readiness Questions
Before spending on a humanoid platform, answer these five questions honestly. For most teams, the honest answer to at least three of them will be "no", and that is fine. It means you should invest in robot arms and teleoperation infrastructure first.
- Does your research specifically require bipedal locomotion combined with manipulation? If your manipulation tasks can be done on a tabletop, a robot arm will collect better data faster, for a fraction of the price of a humanoid. Humanoids add locomotion complexity that is irrelevant to most manipulation research.
- Do you have a lab space with flat floors, a clear test area and safety infrastructure? Humanoids fall. When a 30 to 70 kg robot falls, it can damage itself, the room and anyone nearby. You need padded flooring or a clear perimeter, a way to suspend the robot for testing, and emergency stops everyone can reach.
- Do you have at least one engineer with experience in locomotion control or whole-body control? Humanoid software stacks are less mature than arm manipulation frameworks. Expect to spend real time on balance, whole-body kinematics and coordinating walking with manipulation.
- Is your budget sufficient for the full system, not just the robot? Budget separately for hands, onboard compute, spare batteries, a safety rig, teleoperation equipment, a training workstation and integration engineering time. See total cost of ownership below.
- Have you already published results or built products with simpler robots? If your team has not yet mastered single-arm manipulation, starting with a humanoid is like learning to drive in a Formula 1 car. The fundamental algorithms (perception, grasping, planning) are the same, but debugging them on a humanoid is much harder and slower.
If you answered "yes" to at least three of these questions, a humanoid purchase makes sense. If not, start with a robot arm: you will make faster progress and spend less money. Come back to humanoids when your manipulation foundations are solid.
Humanoid Robot Comparison Table: The Leading Platforms
Humanoids you can buy today
Prices are live from our US store. "From" means the listing has more than one configuration. Specs are the manufacturers' own: see the linked guides for sources.
| Platform | Price | Height / weight | DoF | Development access | Best for |
|---|---|---|---|---|---|
| Unitree G1 | Base $21,600.00; EDU from $43,900.00 | 132 cm / ≈35 kg | 23 (base), up to 43 (EDU) | EDU only | Research, teleoperation data, learning-based control |
| Unitree H1 | $99,900.00 | ≈180 cm / ≈47 kg | 5 per leg, 4 per arm | Yes | Full-size locomotion research |
| Unitree H1-2 | $128,900.00 | ≈178 cm / ≈70 kg | 27 | Yes | Full-size loco-manipulation |
| Unitree H2 | From $40,900.00 | 182 cm / ≈70 kg | 31 | EDU only | Full-size research (EDU), demos and reception (base) |
| Booster T1 | Request a Quote | ≈118 cm / ≈30 kg | 23 to 41 | Yes | Locomotion and RoboCup research |
| Booster K1 | $12,000.00 | ≈95 cm / ≈19.5 kg | 22 | Yes | Education, competitions, first projects |
| Unitree R1 | $6,370.00 | ≈123 cm / ≈29 kg | 20 to 40 | EDU only | Entry-level demos and teaching |
| AGIBOT X2 | From $27,999.00 | 131 cm | Varies by model | X2 Ultra | Interaction, performances, education |
For every humanoid we sell, including AGIBOT's full-size models, Fourier, EngineAI, LimX and others, see the humanoid price guide. To put any of them side by side, open the humanoid comparison.
Humanoids you cannot simply buy
| Platform | How it is offered (October 2026) |
|---|---|
| Tesla Optimus | Not on public sale. What Tesla has said about price and timing is collected in Tesla Optimus: price and availability |
| Figure 03 | Presented on figure.ai as a home humanoid, with no order page or published price |
| Boston Dynamics Atlas | Offered to enterprise customers; Boston Dynamics lists it at 1.9 m and 90 kg, with a 4-hour battery, and asks prospects to get in touch |
| Agility Digit | Deployed with industrial customers through Agility's own deployment process; assembled at Agility's RoboFab in Salem, Oregon (Agility) |
| 1X NEO | A home robot; 1X takes orders with a $200 deposit |
Deep Dives: Three Platforms You Should Understand
Unitree G1: The Most Accessible Humanoid
The Unitree G1 is the humanoid most research labs start with. Unitree sells it in two versions that look the same but are very different purchases. The base G1 (current store price: $21,600.00) has 23 degrees of freedom and, per Unitree's spec table, no secondary development. The G1 EDU (from $43,900.00 in our store) is open to secondary development, goes up to 43 degrees of freedom with optional dexterous hands and extra waist and wrist joints, takes an added compute module, and carries an 18-month warranty instead of 8. If you will write code, you need an EDU. Every configuration is priced in the G1 guide.
What works well: walking on flat floors is smooth and reliable. Unitree's open SDK, ROS 2 packages, RL training code and XR teleoperation code all support the G1, and its community is the largest of any humanoid. What does not: at 132 cm it is compact, so it cannot work at standard counters and shelves the way an adult can. Unitree rates arm load at about 2 kg (3 kg on the EDU), which limits manipulation to light objects. The battery lasts about 2 hours, so long data sessions need spare packs. Dexterous hands add capability, and also cost and failure modes. For labs starting humanoid research on a reasonable budget, the G1 EDU is the default choice in 2026.
Booster T1: The Locomotion Specialist
The Booster T1 (Request a Quote) is a 1.2 m, 30 kg humanoid that Booster Robotics calls the 2025 RoboCup "AdultSize" champion model. It is open to development, and Booster publishes its SDK, robot models and RL training and deployment code on GitHub. Choose it for dynamic locomotion, soccer and RL research. Its base configuration has 4-DoF arms, so manipulation work needs the gripper or dexterous-hand version, or Booster's larger 31-DoF T2.
Unitree H1 and H2: Full Size
When the task needs adult height and reach, the choice is between Unitree's H1, H1-2 and H2. The H1 ($99,900.00) is the fastest and lightest of the three and carries a 3D LiDAR and depth camera. The H1-2 adds 7-DoF arms. The H2 is Unitree's newest full-size body, with 31 degrees of freedom; as with the G1, only the H2 EDU is open to secondary development. Full-size humanoids need more space, a stronger safety rig and more people around every test than a G1 does.
Humanoid vs Arm vs Quadruped: Which Should You Buy?
This is the most important decision in this guide. Many teams default to humanoids because of the excitement around them, but for most use cases a different form factor is more practical and cost-effective.
| Your Need | Best Form Factor | Recommended Platform |
|---|---|---|
| Tabletop manipulation research | Robot arm | OpenArm$5,400.00 or Franka FR3 |
| Mobile inspection / patrol | Quadruped | Unitree Go2$3,790.00 or Boston Dynamics SpotRequest a Quote |
| Mobile manipulation (navigate + pick) | Quadruped + arm | Unitree B2$85,900.00 + Z1 arm$20,799.00 |
| Two-armed tasks around a building (kitchens, cabinets, doors) | Wheeled bimanual | RC Mobile Aloha$20,000.00 |
| Bipedal locomotion research | Humanoid | Unitree G1 EDU$43,900.00 or Booster T1Request a Quote |
| Loco-manipulation research | Humanoid | Unitree G1 EDU$43,900.00 or Unitree H1-2$128,900.00 |
| Human-scale environment tasks | Humanoid | Unitree H1$99,900.00 or Unitree H2 EDU$66,800.00 |
| Teaching and outreach | Small humanoid | Booster K1$12,000.00 or Unitree R1$6,370.00 |
If your tasks need two arms and the ability to move between rooms, but not legs, a wheeled bimanual platform is the cheaper and safer middle ground. Stanford's Mobile ALOHA puts the two-arm ALOHA teleoperation setup on a mobile base. Its authors report that with 50 demonstrations per task, co-training with existing static ALOHA data raised success rates by up to 90%, on tasks such as opening a two-door wall cabinet to store heavy pots and calling and entering an elevator. A legged humanoid collecting the same demonstrations also needs whole-body teleoperation, where the operator's body motion is retargeted onto the robot's joints, which is slower to learn and to run.
The key insight: for most manipulation tasks, a robot arm on a fixed base is more reliable, more repeatable, cheaper, and easier to program than a humanoid. Humanoids only make sense when you specifically need the combination of bipedal locomotion and upper-body manipulation, or when the environment was designed for humans (stairs, narrow corridors, human-height workstations) and cannot be modified. The quadruped guide covers robot dogs by use case.
Real Use Cases for Humanoids in 2026
Setting aside demo videos and marketing claims, here is what humanoid robots are actually doing today:
Teleoperation data collection: This is the primary use case for humanoid purchases in 2026. Research labs buy humanoids to collect whole-body manipulation demonstrations by teleoperation, then train autonomous policies from that data. The Unitree G1 is the most popular platform for this work because of its price and open software. Our data services team collects humanoid demonstrations for clients who need the data but do not want to own and operate the hardware.
Locomotion and gait research: University labs studying bipedal locomotion, balance and terrain use humanoids to test algorithms trained in simulation. The sim-to-real gap for locomotion is significant, so physical hardware is essential for validating RL-trained gaits. Unitree and Booster both publish open RL pipelines that go from simulation to the real robot.
Controlled industrial pilots: Agility Digit, Boston Dynamics Atlas and wheeled humanoids such as the AGIBOT G2 are offered for material handling and manufacturing work. These are engineered deployments with vendor support on specific, repetitive tasks, not general-purpose autonomous workers.
Events, reception and education: Humanoids that walk, gesture and talk draw crowds at trade shows, lobbies and classrooms. Small, robust robots such as the Booster K1 and interactive ones such as the AGIBOT X2 are built for this. See humanoid robots by use case.
Human-robot interaction research: Some labs use humanoids to study human comfort, trust and collaboration with human-form robots, which matters for applications in healthcare, retail and public spaces.
Honest Capability Assessment: What Works and What Does Not
This table reflects our assessment from operating Unitree G1 units and reviewing published results from other platforms. Your experience will vary with the platform and your team's expertise.
| Task | Status | Notes |
|---|---|---|
| Walking on flat surfaces | Works reliably | All major platforms walk smoothly on flat, hard floors. |
| Pick and place (known objects) | Works sometimes | Reliable in structured setups. Degrades with novel objects, varied lighting, or cluttered scenes. |
| Stair climbing | Works sometimes | Demonstrated, but not reliable enough for unsupervised use. Fall risk is significant. |
| Carrying objects while walking | Works sometimes | Purpose-built industrial platforms do this for totes. Smaller research humanoids struggle with balance under load. |
| Manipulation with tools | Not yet | Demonstrated in curated demos. Not reproducible in unstructured settings on any platform. |
| Language command following | Works sometimes | Voice commands for simple moves work on interactive robots. Reliability on open-ended tasks is low. |
| Fall recovery | Mixed | Some small robots, such as the Booster K1, get up on their own after a fall. Full-size humanoids generally need a person and a hoist. |
| Multi-step autonomous tasks | Not yet | Chaining walk + navigate + grasp + place autonomously is beyond current reliable capability. |
Total Cost of Ownership
The purchase price of a humanoid is only part of what you will spend in the first year. Here is what to budget for on top of the robot, with live prices for the items we stock:
- Development access. If you will write code, price the EDU configuration, not the base model. For the G1 that is the difference between $21,600.00 and $43,900.00 for the entry EDU.
- Hands. Many humanoids ship with simple or dummy hands. Dexterous hands are priced separately; for the H2, a Dex3-1 three-finger hand is $6,000.00 each.
- Onboard compute for running learned policies, such as an Orin backplane dock for the G1 ($7,600.00).
- Spare batteries and chargers. Plan sessions around 1 to 3 hours per pack.
- Safety infrastructure: padded flooring, a perimeter, emergency stops, and a hoist or bracket for full-size robots, such as the H2 protective bracket ($1,080.00).
- Teleoperation equipment: an XR headset or leader arms, plus external cameras for third-person views.
- A training workstation with a current NVIDIA GPU for simulation and policy training.
- Spare parts (joint modules, cables, connectors) and a support plan.
- Integration engineering time, usually the largest cost and the one nobody writes down.
Before you buy, ask every vendor three questions: where are spare actuators physically stocked, what is the warranty response time in business days, and what is the real repair turnaround if a joint fails? Clear answers are a good sign.
How to Lease Instead of Buy
For many teams evaluating humanoids, leasing is more sensible than purchasing. The technology is moving fast: the platform you buy today may be superseded by a significantly better model within 18 months. Leasing preserves capital, reduces risk, and lets you upgrade when better hardware arrives.
We lease humanoids for research and enterprise teams on monthly, quarterly and annual terms, with a one-month minimum and setup support included. Terms are quoted per project. For current availability, visit our leasing page or contact us directly.
Getting Started: Your First 90 Days with a Humanoid
If you have decided to purchase or lease a humanoid, here is a realistic timeline for your first three months. This assumes a Unitree G1 EDU (the most common starting platform) and a team with ROS 2 experience.
Days 1-14: Setup and safety. Unbox and inspect the hardware. Set up the safety infrastructure (padded area, barriers, emergency stops, a way to suspend the robot). Install the Unitree SDK and ROS 2 packages on your workstation. Verify communication with the robot and update firmware. Do not attempt walking until the safety infrastructure is complete. Budget a full week for this: rushing setup leads to falls and hardware damage.
Days 15-30: Basic locomotion. Run the built-in walking controller on flat ground inside your safety area. Learn the remote controller. Practise emergency stops until they are reflexive. Log IMU and joint data to verify sensor quality. At the end of week 4, you should be able to walk the robot reliably in your lab.
Days 31-60: Arm manipulation. Switch focus to upper-body control. Set up teleoperation (XR headset or leader arms). Collect first demonstrations for simple tasks (reaching, grasping light objects). Learn the arms' joint limits and workspace empirically. Build your data pipeline: camera calibration, recording and episode management.
Days 61-90: Integration and first experiments. Combine locomotion and manipulation in simple scenarios (walk to a table, pick up an object). Run your first policy training experiments on the demonstrations you collected. Identify the bottlenecks in your pipeline, which are usually data quality rather than algorithm choice. By day 90 you should have a working data-collection pipeline and preliminary policy results, or a clear understanding of why you need a different platform.
A candid note: most teams underestimate the time required for the first 90 days. Hardware debugging, driver issues, safety procedures and the learning curve of operating a bipedal robot all take longer than expected. Plan for it and protect your team's time during this period.
Frequently Asked Questions
How much does a humanoid robot cost?
In our store, from $6,370.00 for the Unitree R1 and $12,000.00 for the Booster K1, through $21,600.00 for the base Unitree G1, to $99,900.00 for the full-size Unitree H1. Research configurations with development access, hands and compute cost more. Every model is priced in the humanoid price guide. Total cost of ownership, including hands, compute, safety, teleoperation and engineering time, is well above the robot's price in the first year.
Can I buy a Tesla Optimus?
No. Tesla Optimus is not on public sale. What Tesla has said about price and timing is collected in Tesla Optimus: price and availability. Plan your work around platforms you can buy today.
Is Unitree G1 good for research?
Yes, in its EDU configuration. It has the largest community of any humanoid, open SDK, ROS 2, RL and teleoperation code, and EDU configurations take dexterous hands for up to 43 degrees of freedom. The main limitations are its compact size (132 cm), light arm load (about 2 to 3 kg) and roughly 2-hour battery. Buy an EDU: Unitree lists secondary development for the EDU only.
What is the best humanoid for data collection?
The Unitree G1 EDU is the best value for teleoperation data collection: Unitree's open xr_teleoperate supports it with Apple Vision Pro, PICO 4 Ultra Enterprise and Meta Quest 3, and its size is manageable in a lab. For human-scale reach, the Unitree H1-2$128,900.00 and EDU-class H2 are the step up. We also collect humanoid teleoperation data to order if you need demonstrations without owning the hardware.
Do humanoid robots work in real environments yet?
In controlled, structured environments, partially. Industrial humanoids handle specific material-handling tasks in engineered deployments with vendor support. In unstructured environments (homes, public spaces, varied terrain), humanoid robots do not yet work reliably without supervision. If someone tells you otherwise, ask to see their failure-rate data, not their highlight reel.
Should I buy a humanoid or a robot arm?
For most teams, buy a robot arm. An OpenArm$5,400.00 or Franka FR3 will produce manipulation research results faster, cheaper and more reliably than any humanoid. Buy a humanoid only if your research specifically requires bipedal locomotion combined with manipulation, or if your application requires the human form. See the decision table above.
How long does a humanoid robot battery last?
Manufacturers quote about 2 hours for the Unitree G1 and Booster T1 when walking, and about 3 hours for the Unitree H2. In practice, research sessions are often shorter because of thermal limits, safety supervision and operator fatigue during teleoperation. Buy spare batteries and plan data collection in blocks with time for battery swaps.
Is a Chinese humanoid affected by the FCC Covered List?
On July 28, 2026 the FCC added foreign-produced "advanced robotic devices", which it defines as mobile robots such as humanoids and quadrupeds, to its Covered List. According to the FCC fact sheet, the update "does not prohibit the import, sale, or use of any existing models" the FCC previously authorized, and it "does not affect any previously purchased devices." New models generally cannot receive FCC authorization. If compliance matters for your organization, ask us for the FCC ID of the exact configuration before you order.
Lease or Buy a Humanoid Through Robotics Center
Whether you want to buy a humanoid for your lab, lease one for a project, or have us collect humanoid teleoperation data for you, we can help. Talk to our team about your use case and timeline.