Humanoid Robot Hands – State of the Art
The hand decides what a humanoid can actually do. No robot hand has yet proven it can hold up to real work, day after day. This is the state of the art: 20 chapters, 574 references and a card for every one of 78 hands.
The robot hand market, from tendon to term sheet
The headline numbers all point up. GGII expects about 70,200 dexterous hands to sell in China in 2026, up from about 19,200 in 2025. ITjuzi counts about $4.2 billion of disclosed capital in China’s dexterous-hand track between January 1 and August 3 – more than in all of 2025, though about two-thirds of it was whole-robot rounds at humanoid makers that build their own hands. Around Beijing’s World Robot Conference, high-DOF hands that once sold for RMB 400,000–500,000 were quoted at the RMB 10,000 level. Yet the best published system-level runs for any humanoid hand still fall 8 to 50 times short of the 10,000-hour MTBF that Xiaomi founder Lei Jun proposed as a 2027 target for robots in specific industrial scenarios.
This report maps the whole field at our September 5, 2026 cutoff: who makes hands and who builds their own, how tendons, motors at the joint and hydraulics really compare, what touch and force sensing are worth, how long hands last, what they cost to build and to own, where the money goes – and a catalog card with a Hand Score for every one of the 78 hands in the humanoid.guide registry.
- 325
- pages, 35 of them references
- ~81,000
- words of analysis, plus ~24,000 in the references
- 574
- references, from patents and vendor data to academic papers
- 20
- chapters, plus an Executive Summary and 10 appendices
- 45
- tables, plus 8 figures
- 78
- hand cards, each with a Hand Score
how far the best published system-level hand runs fall short of the 10,000-hour MTBF Xiaomi’s Lei Jun proposed as a 2027 target for robots in specific industrial scenarios
disclosed capital in China’s dexterous-hand track, January 1 – August 3, 2026 (ITjuzi) – above all of 2025, about two-thirds of it whole-robot rounds
dexterous hands GGII expects to be sold in China in 2026, up from about 19,200 in 2025
the level high-DOF hands were quoted at around WRC 2026 – down from RMB 400,000–500,000
Get the free preview
A 47-page walkthrough of what every chapter covers and which questions it answers, with the pages of every chapter at a glance. Download it right away – we email it too.
One email. Unsubscribe any time.
Six desks, one hand
The hand is where hardware, data, money and procurement meet. The report is organized around the questions these readers ask most.
Hand and component vendors
Where your product sits among 78 cataloged hands, which business templates have evidence behind them, and the diligence questions buyers and investors will put to you.
Humanoid OEMs
Build or buy: who makes their own hands, a ranked sourcing framework, and the fleet volume at which building stops losing to buying.
Buyers and integrators
Three-year cost of ownership worked through for four hands, an RFP checklist to print into a tender, and what to demand on durability before you sign.
Investors
The two-camp financing structure, a unit-economics model of a hands-only vendor, valuation frameworks and the exit routes.
Research labs
A selection guide that matches hand platforms to research programs, the sim-to-real toolchain and the benchmark landscape.
Strategy and policy
Regional ecosystems, the patent landscape, the certification gap and scenarios for the hand market to 2031.
Hands have never been more capable. Nobody has yet shown one that lasts.
The best published system-level runs for any humanoid hand – 200 to more than 1,250 hours – fall 8 to 50 times short of the 10,000-hour MTBF that Xiaomi founder Lei Jun proposed as a 2027 target for robots in specific industrial scenarios. They are run times, not failure statistics, and none is long enough to prove that target even with zero failures. At our cutoff no standard defined a hand-durability test. The report shows where hands fail, what wear costs per operating hour, and what to demand from a vendor before you buy.
Six findings define the state of the art
Reliability is the binding constraint
No field data published since has contradicted Roland Berger’s assessment that advanced hands last less than a year in volume use.
Motors at the joint are gaining
Tendons, direct drive and gearing still coexist, but 53% of the registry hands that disclose a drive now put motors at the joint. Hybrids, not convergence.
The price war reached hands
Around WRC 2026, high-DOF hands that once sold for RMB 400,000–500,000 were quoted at the RMB 10,000 level.
The money splits into two camps
OEMs that build their own hands take most of the money, in whole-robot rounds. Independent suppliers fight for the merchant market – and capital is concentrating in fewer, larger names.
Four templates, one trap
Module supply, components and data pipelines have evidence behind them; licensing is unproven. The undifferentiated complete-hand vendor is the trap.
Data is the second bottleneck
Policies ship in months, hands in years. Vendors that sell the data pipeline alongside the hand are, in our view, pricing the bottleneck correctly.
Free · the hand registry
Compare the hands for yourself
Specs, prices and Hand Scores for every hand in the humanoid.guide registry – the catalog this report starts from, now including hands added since our cutoff.
Twenty chapters, ten appendices and an Executive Summary
- 00
Executive Summary
The central argument, six findings, and what would change our view.
- 01
Introduction: Scope, Method, and the Humanoid Guide Hand Score
Why the hand decides a humanoid’s value, how to read a vendor datasheet, and what the Hand Score can and can’t tell you.
- 02
Market Landscape: The Registry, the Tiers, and the Hand Score
The 81-entry registry (78 distinct hands), four price tiers, shipments, price–performance since 2020 and a 2026 demand model.
- 03
The Hands-Only Vendors: Who They Are and What Makes Each Different
LinkerBot, INSPIRE, AGILINK, Xynova, PaXini, Sharpa, Shadow, the Korean tier and the Western challengers.
- 04
Humanoid OEMs and Their Hands: Who Builds, Who Buys
Tesla, Figure, 1X and the Chinese OEMs: build or buy, a sourcing framework and the real customers of the merchant market.
- 05
Actuation I – Tendon-Driven Systems
Why tendons win the design review, what wears them out, and three case studies from Shadow to DexHand021.
- 06
Actuation II – Motors at the Joint: Direct Drive, Linkage, and the Gearing Stack
Reducers, roller screws, coreless motors, the thermal ceiling, a worked actuator sizing and the thumb problem.
- 07
Actuation III – Hydraulic, Soft, and Exotic Approaches
Sanctuary’s hydraulics, soft robotics, lab-grown muscle – and why electromechanics keeps winning.
- 08
Haptics and Tactile Sensing
Six transduction families, the bandwidth wall, coverage strategies and skin as a consumable.
- 09
Force Sensing and Control
Four force-sensing modalities, control architectures, slip, and a worked real-time budget.
- 10
Reliability and Durability: The Bottleneck Quantified
The published record, a failure-mode analysis, maintenance economics and how to believe a cycle claim.
- 11
Manufacturing and Supply Chain
Two bills of materials, China’s clusters, the 2026 price war and learning curves.
- 12
AI, Data, and the Hand
Sim-to-real, manipulation foundation models, the data-collection economy and the simulators compared.
- 13
Financing and Business Models
The 2026 financing picture, four business templates, a unit-economics model, valuation and exits.
- 14
Bottlenecks and Challenges: The Synthesis
Eight constraints rated on severity and tractability, five propositions tested, and how the bottlenecks bind together.
- 15
Standards, Safety, and Certification
ISO 10218, collaborative force limits, China’s standards work and the certification gap.
- 16
Patent Landscape and IP Strategy
Where the filings concentrate, why the thicket hasn’t caught fire, and playbooks by vendor class.
- 17
Regional Ecosystems and Supply Clusters
China, the United States, Europe, Korea and Japan, layer by layer, and the policy instruments behind them.
- 18
Procurement, Integration, and Total Cost of Ownership
The real cost stack, four hands compared over three years, an RFP checklist and an illustrative 14-week pilot.
- 19
Outlook: The Next Twelve Months and the Road to 2031
A watch list to Q3 2027, scenarios to 2031, recommendations and the wildcards.
- 20
Conclusion
What the industry has achieved, what it hasn’t, and what we will track.
- A–J
Appendices
Hand Score method, vendor and component-supplier directories, extended spec tables, glossary, benchmarks, a failure-mode library, a researcher’s selection guide and the 78-card Hand Catalog.
45 tables. 8 figures. 78 hand cards.
The Hand Catalog: a card for every hand in the registry at the cutoff, each with its Hand Score.
System-level operating evidence set against the 5,000- and 10,000-hour thresholds.
A worked cost-of-ownership comparison over three years at 4,000 hours a year.
Where the money goes in a ~$7,500 in-hand hand and a ~$70,000 tendon hand – our bottom-up estimates.
The bottleneck matrix: eight constraints rated on severity and tractability.
Dexterity against price across commercial hands, September 2026.
Two licenses. One-time purchase.
- Full 325-page report (PDF)
- Executive Summary, 20 chapters and 10 appendices
- 45 tables, 8 figures, 574 references
- The 78-card Hand Catalog with Hand Scores
- Single-user license
- Free updates throughout the 2026 cycle
- Email support
- Everything in Single User
- Unlimited internal users at one organization
- Right to quote in internal engineering, sourcing and investment documents
- Priority email support
- Optional 60-minute briefing call
Single User License · Terms · Enterprise License · Terms
By purchasing, you agree to the terms of the license you select. Questions? human@humanoid.guide
Frequently asked questions
Who is the report written for?
Hand and component vendors, humanoid OEMs, buyers and integrators, investors, research labs choosing a platform, and strategy and policy teams. It defines every convention it uses, so you don’t need a background in hand design – but it doesn’t simplify the engineering.
How is this different from the Humanoid Actuation Report?
The Actuation Report takes the whole humanoid joint apart and gives dexterous-hand actuation one chapter. This report is about the hand alone: vendors, OEM sourcing, tendons versus motors at the joint, touch and force sensing, durability, manufacturing, financing, standards, patents and procurement – plus a catalog card for every hand in the registry at our cutoff.
What is the Hand Score?
A rating that combines a hand’s degrees of freedom and its strength into one number, so hands can be screened quickly. Appendix A publishes the scoring bands, and Chapter 2 explains where the score misleads: it is a screen, not a ranking.
Where do the numbers come from?
574 references: manufacturer disclosures, patent filings, market research, financial reporting, academic literature, product registries and press coverage. Vendor claims we could not verify are marked as vendor claims, our own estimates are marked as ours, and every price follows one stated rule – the vendor’s own price, otherwise a dated reseller price, otherwise our estimate, marked “(est.)”. Prices after a subsidy say so, and the Hand Catalog marks the few registry-listing prices it uses “(listing)”.
How current is it, and will it be updated?
The data runs to September 5, 2026, with reseller prices checked in October 2026; the few later developments we mention are dated and marked as such. Buyers receive free updates throughout the 2026 cycle.
Can I see a preview first?
Yes – the free preview walks through every chapter and shows the report’s pages at low resolution.
Do you do custom research?
Yes – vendor deep dives, hand benchmarking, sourcing support and internal briefings. Get in touch to discuss scope.
The hand decides what a humanoid can do
In 2026 the hand became the busiest corner of humanoid robotics: more vendors, more capital, falling prices – and still no proof of durability. This report reads it chapter by chapter, with a card for every hand.
Buy full report · from $590