Humanoid Robot Buyer's Guide 2026: Compare Models, Costs & Use Cases

The State of Humanoid Robots in 2026

The humanoid robot market in 2026 is characterized by enormous hype, massive venture capital investment, and a growing gap between demo videos and real-world capability. Over $6 billion has been invested in humanoid startups since 2023. At least eight companies have full-body prototypes walking and manipulating objects. But here is the honest assessment: only two platforms are available for purchase by external customers today (Unitree G1 and H1), one is available for enterprise lease (Agility Digit), and the rest are either in closed partner programs, internal deployment, or pre-commercial development.

What works: bipedal walking on flat surfaces is reliable across most platforms. Basic pick-and-place manipulation (known objects, structured environments) works on the platforms with decent arms. Teleoperation — a human operator controlling the humanoid remotely — works and is the primary method for collecting training data for autonomous behavior.

What does not work yet: autonomous manipulation of novel objects, recovery from falls or unexpected perturbations, operation in genuinely unstructured environments, and any task requiring multi-step reasoning 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 or a data collection tool, not an autonomous worker.

Are You Ready for a Humanoid? Five Readiness Questions

Before spending $16,000-250,000 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.

1. Does your research specifically require bipedal locomotion combined with manipulation? If your manipulation tasks can be done on a tabletop, a $3,500 robot arm will collect better data faster than a $16,000 humanoid. Humanoids add locomotion complexity that is irrelevant to most manipulation research.
2. Do you have a lab space with flat floors, at least 5m x 5m of clear area, and safety infrastructure? Humanoids fall. When a 35-90 kg robot falls, it can damage itself, the environment, and nearby humans. You need padded flooring, safety barriers, emergency stop systems, and space for the robot to move without collisions. Most university labs do not have this without modification.
3. Do you have at least one engineer with experience in locomotion control or whole-body control? Humanoid software stacks are immature compared to arm manipulation frameworks. You will spend significant time on gait tuning, balance recovery, whole-body inverse kinematics, and locomotion-manipulation coordination. Without locomotion expertise, you will spend months on problems that an arm-based lab would skip entirely.
4. Is your budget sufficient for the full system, not just the robot? The hardware is 40-60% of total cost. Budget separately for: safety infrastructure ($5,000-15,000), spare parts and warranty extension ($3,000-10,000/year), dedicated compute with GPU ($3,000-8,000), teleoperation equipment ($2,000-10,000), and integration engineering time (2-6 person-months).
5. 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 5-10x 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 8 Leading Platforms

This table covers every humanoid platform with meaningful traction as of April 2026. Prices are for the base unit where available. "Not for sale" means exactly that — no amount of money will get you one through normal channels.

Deep Dives: Three Platforms You Should Understand

Unitree G1: The Most Accessible Humanoid

The Unitree G1 at $16,000 is, without exaggeration, the most important product launch in the humanoid robot market. Before the G1, the cheapest humanoid platform available for research purchase was $90,000+ (the Unitree H1 itself or the Fourier GR-1). The G1 dropped that floor by 80%, opening humanoid research to labs that previously could only afford robot arms. The base configuration includes 23 DOF for locomotion and basic arm manipulation. The EDU variant adds dexterous hands for a total of 43 DOF at approximately $28,000.

What works well on the G1: walking on flat surfaces at up to 2 m/s is reliable and smooth. The Python SDK and ROS2 driver (unitree_ros2) are functional and actively maintained. Teleoperation for data collection works via VR controllers or leader-follower setups. The community is the largest of any humanoid platform, with active GitHub repos, Discord servers, and university collaborations. What does not work well: the 1.27 m height means the G1 cannot reach standard countertops or shelves — it is a compact humanoid, closer to a child's height. The 2 kg per-arm payload limits manipulation to lightweight objects. Battery life is 1-2 hours under active use, making long data collection sessions require battery swaps. The dexterous hand option adds capability but also significant complexity and failure modes. For labs starting humanoid research on a reasonable budget, the G1 is the clear first choice in 2026.

Figure 02: The Most Hyped Platform

Figure AI has raised over $2.6 billion, making it the most well-funded humanoid startup in history. The Figure 02 is their second-generation platform, designed for commercial deployment in partnership with BMW and other manufacturing companies. Figure's key differentiator is their hand design — the 02's hands are among the most dexterous in the industry, with individual finger actuation that enables manipulation tasks other humanoids cannot attempt.

The honest assessment: you cannot buy a Figure 02. The company operates a partner program model — they deploy units to partner facilities (primarily BMW's Spartanburg plant) with Figure engineers on-site. There is no self-service purchase option, no SDK documentation available publicly, and no timeline for general availability. The demo videos are impressive but heavily curated. If you are a large manufacturer with an existing relationship with Figure AI, you may be able to join their partner program. For everyone else, the Figure 02 is something to watch, not something to plan around. If your research timeline requires hardware in the next 12 months, look at Unitree or Agility instead.

Agility Digit: The Most Deployed Humanoid

Agility Robotics' Digit is the humanoid with the most validated real-world deployment hours. Their partnership with Amazon for warehouse tote handling has produced more operational data than any other humanoid program. Digit's design philosophy is pragmatic: the legs use a unique reverse-knee configuration optimized for walking efficiency over aesthetics, and the arms are designed for carrying and placing — not for dexterous manipulation.

Digit is available as an enterprise lease at approximately $250,000, which includes hardware, a support contract, software updates, and on-site deployment assistance. The target customer is large logistics and manufacturing companies deploying 10+ units. Agility operates a factory in Salem, Oregon (RoboFab) with capacity to produce thousands of units per year. For research labs, Digit is generally not accessible — the lease terms and pricing are designed for enterprise fleet deployments, not individual unit research. If your use case is warehouse logistics at scale and you have the budget, Digit is the most production-ready humanoid on the market. For everything else, the cost and enterprise focus make it impractical.

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.

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.

Real Use Cases for Humanoids in 2026

Setting aside demo videos and marketing claims, here is what humanoid robots are actually doing today in deployed or near-deployed settings:

Teleoperation data collection: This is the primary use case for humanoid purchases in 2026. Research labs buy humanoids to collect whole-body manipulation demonstrations via teleoperation, then train autonomous policies from that data. The Unitree G1 is the most popular platform for this work because of its price and SDK quality. SVRC operates G1 units for teleoperation data collection for clients who need humanoid demonstrations but do not want to own and operate the hardware.

Locomotion and gait research: University labs studying bipedal locomotion, dynamic balance, and terrain adaptation use humanoids as physical testbeds for algorithms developed in simulation. The sim-to-real transfer gap for locomotion is significant, making physical hardware essential for validating RL-trained gait policies.

Controlled manufacturing pilots: Agility Digit units are operating in Amazon warehouses and Agility's own RoboFab, performing tote-moving tasks in structured environments. Figure 02 units are in BMW's Spartanburg plant. These are controlled pilots with on-site engineering support, not autonomous deployments. The robots handle specific, repetitive tasks in environments that have been modified to accommodate them.

Human-robot interaction research: Some labs use humanoids for studies on human comfort, trust, and collaboration with human-form robots. The humanoid form factor elicits different human responses than arm or wheeled 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 based on operating Unitree G1 units at SVRC and reviewing published results from other platforms. Your experience may vary depending on the specific platform and your team's expertise.

Total Cost of Ownership

The purchase price of a humanoid is 40-60% of what you will actually spend in the first year. Here is a realistic breakdown for the two purchasable platforms:

Unitree G1: First-Year Cost Estimate

• Base unit (23 DOF): $16,000
• Dexterous hand upgrade (EDU, optional): +$12,000
• Safety infrastructure (padded flooring, barriers, e-stop system): $3,000-5,000
• Spare batteries (recommended 2 extra): $1,200
• Compute workstation (GPU for policy training): $3,000-5,000
• Teleoperation equipment (VR headset + controllers or leader arms): $500-10,000
• Cameras (external RealSense or ZED for third-person view): $600-2,000
• Spare parts reserve (servos, cables, connectors): $2,000
• Integration engineering time (2-4 person-months at your labor rate): varies
• Total first year (base): $26,000-32,000
• Total first year (EDU with hands): $38,000-44,000

Unitree H1: First-Year Cost Estimate

• Base unit: $90,000
• Safety infrastructure (larger space, reinforced barriers): $8,000-15,000
• Extended warranty / support contract: $5,000-10,000
• Compute and teleoperation: $5,000-15,000
• Facility modifications (floor reinforcement, power, ventilation): $3,000-10,000
• Integration engineering: 3-6 person-months
• Total first year: $115,000-150,000

How to Lease Instead of Buy

For most teams evaluating humanoids, leasing is more sensible than purchasing. The technology is evolving rapidly — 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.

SVRC offers humanoid leasing for research and enterprise teams:

• Unitree G1 lease: Starting at $1,200/month on 6-month terms. Includes basic support, spare battery, and ROS2 setup assistance. Available for qualified research labs and companies.
• Unitree G1 EDU (with hands) lease: Starting at $2,200/month on 6-month terms. Includes teleoperation setup and operator training session.
• Short-term rental: G1 units available for 1-3 month project-based rentals at $1,800/month. Ideal for evaluating whether a humanoid fits your research before committing to a purchase or longer lease.

All leases include SVRC engineering support for initial setup, ROS2 driver configuration, and teleoperation pipeline integration. For current availability and custom terms, 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 the Unitree G1 (the most common starting platform) and a team with ROS2 experience.

Days 1-14: Setup and safety. Unbox and inspect hardware. Set up safety infrastructure (padded area, barriers, e-stop). Install the Unitree SDK and ROS2 driver on your workstation. Verify communication with the robot. Run factory diagnostics. Do not attempt walking until 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 within your safety area. Learn the joystick control interface. Practice emergency stops until they are reflexive. Log IMU and joint data to verify sensor quality. Start tuning gait parameters for your floor surface. At the end of week 4, you should be able to walk the robot reliably in your lab space.

Days 31-60: Arm manipulation. Switch focus to upper-body control. Set up teleoperation (VR or leader arms). Collect initial manipulation demonstrations for simple tasks (reaching, grasping lightweight objects). Learn the joint limits and workspace boundaries of the arms empirically — they are different from the spec sheet in practice. Start building your data collection pipeline (camera calibration, data recording, episode management).

Days 61-90: Integration and first experiments. Combine locomotion and manipulation in simple scenarios (walk to table, pick up object). Run your first policy training experiments on collected demonstration data. Identify the bottlenecks in your pipeline (usually data quality, not algorithm choice). By day 90, you should have a functioning 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 by 50-100%. Hardware debugging, driver issues, safety procedures, and the learning curve of operating a bipedal robot all take longer than expected. Plan accordingly and protect your team's time from other commitments during this period.

Frequently Asked Questions

How much does a humanoid robot cost?

The Unitree G1 starts at $16,000 (base, 23 DOF) and $28,000 (EDU with dexterous hands). The Unitree H1 is $90,000. Agility Digit leases for approximately $250,000. Other platforms (Figure 02, Atlas, Optimus, Apollo) are not available for individual purchase. Total cost of ownership including support, safety, compute, and integration is typically 2-3x the hardware price in the first year.

Can I buy a Tesla Optimus?

No. As of April 2026, Tesla Optimus Gen 2 is deployed only within Tesla's own factories. Tesla has publicly stated a long-term target price of $20,000-30,000, but no external customer has received a unit, and no purchase timeline has been announced. Plan your research around platforms that are available today.

Is Unitree G1 good for research?

Yes — it is the best humanoid for research in terms of accessibility, price, and community support. The Python SDK and ROS2 driver are functional, the 43-DOF EDU configuration includes dexterous hands, and the growing community means you can find answers to common setup issues quickly. The main limitations are the compact size (1.27 m), low arm payload (2 kg), and 1-2 hour battery life. For labs starting humanoid research, the G1 is the default recommendation.

What is the best humanoid for data collection?

The Unitree G1 is the best value for teleoperation data collection. Its SDK supports real-time joint state streaming at 50+ Hz, it accepts VR controller or leader-arm teleoperation inputs, and its compact size makes it manageable in lab environments. For data collection requiring human-scale reach and heavier payload, the H1 ($90,000) is the step up. SVRC operates G1 units for teleoperation data collection if you need demonstrations without owning the hardware.

Do humanoid robots work in real environments yet?

In controlled, structured environments — yes, partially. Agility Digit handles tote movement in Amazon warehouses. Figure 02 performs specific tasks at BMW. But these are highly engineered deployments with on-site support teams and environments modified for the robots. In unstructured real-world environments (homes, public spaces, varied terrain), humanoid robots do not work reliably. 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. A $3,500 OpenArm or $25,000 Franka FR3 will produce better 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 commercial application requires the human form factor. See the decision table above for specific use case recommendations.

How long does a humanoid robot battery last?

The Unitree G1 and H1 both last 1-2 hours under active locomotion and manipulation. Agility Digit targets 4+ hour shifts using hot-swappable battery packs. In practice, research sessions are often limited to 30-60 minutes due to thermal management, safety supervision requirements, and the cognitive demand on teleoperation operators. Budget for 2-3 spare batteries and plan your data collection sessions around 45-minute active blocks with 15-minute cool-down and battery swap intervals.