humanoid4rent
AnmeldenAngebote einholen

How reliable are humanoid robots? What makers publish on warranty, hours and failures

Humanoid makers publish warranty months and cumulative operating hours, but almost no failure rates. Here is what is on record, what is missing and why uptime claims are hard to compare.

Von Redaktion von humanoid4rent · veröffentlicht 14. November 2026

Diese Seite wurde noch nicht übersetzt
Sie lesen den englischen Text, weil es noch keine Fassung auf Deutsch (Österreich) gibt. Nichts wurde weggelassen: Das ist die ganze Seite, in der Sprache, in der sie geschrieben wurde.

Humanoid robot makers publish little hard reliability data. As of November 2026, none of the companies covered here states a mean time between failures (MTBF), a rated service life or a duty cycle for its humanoid. What you do find is warranty periods (Unitree lists 8 months for the standard G1), total fleet operating hours (Agility reports more than 65,000 for Digit) and occasionally which part failed most often.

This article sets out what is on record, what is missing, and why one maker's uptime claim cannot simply be set against another's.

What the warranty terms say

Warranty is the only reliability-related figure that some makers put directly on a product page, and it is an imperfect proxy. A warranty tells you how long the manufacturer is prepared to pay for defects, not how long the machine is expected to last.

On its G1 product page, Unitree lists a warranty of 8 months for the standard G1 and 18 months for the G1 EDU research version. It also states a battery life of about two hours and a starting price of US$13.5K, excluding tax and shipping. The same page carries an unusual disclaimer: Unitree says the humanoid industry is at an early stage of exploration and advises individual buyers to understand the robots' limitations before purchasing.

What these pages do not say matters as much. None of them defines what counts as a defect versus damage from a fall, whether falls during normal use are covered, what the repair turnaround is, or whether a replacement unit is provided while one is away. Those terms sit in sales contracts that are not public. This article is general information about published figures, not legal advice on warranty or contract terms.

Service plans instead of warranty figures

Boston Dynamics takes a different approach for Spot. Its Spot product page lists an average runtime of 90 minutes, noting that this varies with payload and environment, and an IP54 ingress protection rating against dust and water splashes. It refers buyers to a service programme called Spot Care for long-term support, and says more than 1,500 Spot robots are in customer hands.

The product page itself gives no warranty length, uptime guarantee or service-plan terms; those are handled through sales and the support portal. Reliability, in other words, is sold as a service package rather than published as a number.

Agility Robotics goes further in the same direction. Its customer contract with GXO, according to an Agility press release of 26 October 2026, is a multi-year Robots-as-a-Service (RaaS) agreement. Under such a model the customer pays for the robot's work, and keeping the machine running is largely the supplier's problem. That shifts reliability risk, but it also means the underlying failure data stays inside the supplier.

Operating hours: the figure makers prefer

When makers do publish numbers on reliability, they tend to be cumulative operating hours or work completed.

  • Agility Robotics: in a release on 26 October 2026, Agility said Digit has accumulated more than 65,000 hours of real-world operation across its commercial operations and customer programmes, and has moved more than 100,000 totes at GXO's facility in Flowery Branch, Georgia. It named GXO, AGCO, Toyota Motor Manufacturing Canada, Schaeffler and Mercado Libre as customers.
  • Figure AI: in a post on 19 November 2025 marking the retirement of its Figure 02 fleet, Figure said its robots at BMW logged more than 1,250 hours of runtime on 10-hour weekday shifts, loaded more than 90,000 parts and contributed to the production of more than 30,000 BMW X3 vehicles.

These are real figures from real sites, and they are more than most makers offer. But a fleet total does not tell you how many robots contributed, how often each one stopped, or how long repairs took. Sixty-five thousand hours could mean a handful of robots running for years or many robots running for months. Neither release gives a failure count.

For more on the Digit deployments themselves, see our article on what Agility has reported from warehouse deployments.

The rare account of what actually broke

Figure's BMW post is notable for naming a weak point. It describes the forearm as the top hardware failure point during the BMW deployment, tracing the problem to a circuit board that relayed communications between the main computer and the wrist motors, along with the cabling that had to flex through the joint. Figure says it redesigned the wrist electronics for Figure 03 so that each wrist motor controller talks directly to the main computer, removing that board and the moving cables.

The same post lists targets of more than 99% placement success per shift and zero interventions per shift. It presents these as goals and does not state whether they were met. It also says Figure 02 recorded minimal hardware failures, without giving a number. That is typical: qualitative reassurance, with the quantitative record kept private.

What the research says about field robots

There is no peer-reviewed, independent reliability study of current commercial humanoids that we could find. The closest benchmark is older research on ground robots used in search and rescue and military field work.

In a 2005 paper in IEEE Transactions on Robotics, Carlson and Murphy analysed ten studies covering 15 robot models and found field reliability was low, with a mean time between failures of between 6 and 20 hours. They reported that the main cause differed between studies, with effectors in one and the control system in another, and that limited wireless range and bandwidth were common problems.

An earlier ICRA 2004 paper by Carlson, Murphy and Nelson, covering 15 robots from three manufacturers over several years, reported an overall MTBF of 24 hours and availability of 54%. The authors added an important caution: failures, repair times and downtime varied so widely that standard reliability metrics were poor predictors of future performance.

Those robots were tracked vehicles, not walking humanoids, and the technology has moved on considerably. The findings are useful mainly as a reminder of how much field conditions differ from the lab, and of how hard it is to turn reliability into a single number.

Why uptime claims are hard to compare

Even if every maker published an uptime percentage tomorrow, the numbers would not line up. Several definitions vary from one company to the next:

  1. What counts as a failure. A robot that pauses and needs a person to reset it may be logged as an intervention, not a failure. A fall that the robot recovers from may not be counted at all.
  2. What counts as uptime. Availability can be measured against scheduled shifts, against calendar time, or excluding planned maintenance and charging.
  3. Fleet versus unit. Cumulative fleet hours hide the spread between the best and worst individual robots.
  4. Task and environment. Moving totes on a flat warehouse floor is a narrower, more repeatable job than walking through a crowded exhibition hall with changing lighting and Wi-Fi.
  5. Hardware generation. Figures gathered on one version, such as Digit 4 or Figure 02, do not automatically carry over to its successor.

Agility's own website, for example, notes that Digit 5 is still in development and that certain safety features are not yet complete. Published hours describe the past fleet, not the next model.

What this means for event organisers

Events are a harder setting than most published deployments: short, high-stakes appearances, unfamiliar venues and robots that are transported between jobs. With little public reliability data, the practical answer is to ask direct questions and plan for a stoppage.

  • Ask about the specific unit, not the model: how many hours it has run, when it was last serviced and how old its batteries are. Our guide on battery life and charging during long events covers the runtime side.
  • Ask what happens if it stops. Whether a backup unit or technician is on site matters more than any warranty period. See what happens if the robot fails on the day.
  • Get the fallback in writing. Replacement, partial refund or rescheduling should be stated in the contract; our guide to cancellations and change orders explains what to cover.
  • Compare quotes on the same terms. When suppliers describe reliability differently, set their support, spare units and staffing side by side using how to compare robot hire quotes like for like.
  • Treat warranty months as a floor, not a lifespan. Unitree's 8-month G1 warranty, for instance, tells you about the maker's commitment to defects, not how many events the robot will manage.

Frequently asked questions

Do humanoid robot makers publish a mean time between failures (MTBF)?

Not as of November 2026 for the makers covered here. Unitree, Boston Dynamics, Agility Robotics and Figure AI publish warranty periods, runtime per charge or cumulative operating hours, but none states an MTBF, rated service life or duty cycle for a humanoid. Failure data generally stays with the supplier or appears only in private contracts and service agreements.

How long is the warranty on a Unitree G1?

Unitree's G1 product page lists a warranty of 8 months for the standard G1 and 18 months for the G1 EDU research version. The page does not explain what the warranty covers, for example whether damage from falls is included, or how long repairs take. Those details depend on the sales contract and are not published.

What does Agility's 65,000 hours figure mean?

Agility Robotics said in October 2026 that Digit has accumulated more than 65,000 hours of real-world operation across its commercial operations and customer programmes. It is a fleet total, so it does not show how many robots contributed, how often individual units stopped or how long repairs took. It is evidence of real use rather than a failure rate.

Why can't uptime figures from different robot makers be compared directly?

Makers define failures, interventions and uptime differently. Some count only scheduled shift time, others exclude charging or maintenance, and fleet totals hide differences between individual robots. Tasks also differ: repetitive tote handling in a warehouse is not comparable with walking through a crowded event hall. Figures from one hardware generation do not automatically apply to the next.

What did older research find about robot reliability in the field?

Studies by Carlson and Murphy of search-and-rescue and military ground robots found mean times between failures of 6 to 20 hours in field conditions, and one study reported an overall 24-hour MTBF with 54% availability. These were tracked vehicles rather than humanoids, but the research shows how much harder real environments are than the lab.

How this article was made: written by the humanoid4rent editorial team with AI assistance; every factual statement was checked against the sources listed below on 2026-11-14. Spotted an error? Write to info@humanoid4rent.com.

Quellen

Die Fakten in diesem Artikel wurden am Tag der Erstellung anhand dieser Seiten geprüft.

  1. Unitree G1 Humanoid Agent – Unitree Robotics
  2. Spot – Boston Dynamics
  3. Agility Robotics – Digit humanoid robot
  4. GXO and Agility Robotics Expand Partnership to Accelerate Deployment of Humanoids in U.S. Operations – Agility Robotics
  5. Agility Robotics Highlights Commercial Momentum and Growing Enterprise Demand for Digit Humanoid Robot – Agility Robotics
  6. Production at BMW – Figure AI
  7. How UGVs physically fail in the field – IEEE Transactions on Robotics
  8. Follow-up analysis of mobile robot failures – IEEE ICRA 2004

Mehr aus dem Blog

Möchten Sie bei Ihrem nächsten Event einen Roboter?

Beschreiben Sie einmal Datum, Ort und die Aufgabe des Roboters. Anbieter antworten mit Angeboten, die Sie direkt miteinander vergleichen können.

Anfrage senden

Sie arbeiten in der Robotik? Schreiben Sie für unseren Blog. →