차세대 로봇공학 시장(2027-2037년)

Emerging robotics comprises the classes of machine that break the founding assumptions of classical industrial robotics - operating outside cages, in unstructured environments, alongside people, or beyond the reach of a reliable communications link - and that consequently depend on perception, learning and onboard decision-making rather than on a pre-programmed path in a controlled cell. Emerging robotics is the fastest-growing segment of the global automation economy and the most widely misunderstood. The market spans seven verticals: warehouse and logistics, defence, humanoids, manufacturing and automation, construction and infrastructure, space, and robotics foundation models. Three structural facts define the decade ahead, and each cuts against the prevailing narrative.

The first is that no general-purpose robot exists. Every commercially deployed system in the world today is task-scripted, teleoperated, or supervised. The most-cited humanoid deployments on earth - at BYD, GXO, Amazon, BMW and Mercedes - amount, in total, to a few hundred machines. The growth is real; general autonomy is not. And the systems operating closest to genuine mission-level autonomy are found not in humanoids but in defence, where GPS-denied and communications-denied conditions have forced the problem to be solved rather than deferred.

The second is that value does not sit where capital is flowing. Actuation - actuators together with dexterous hands, which are themselves actuator assemblies - constitutes the overwhelming majority of a humanoid robot's bill of materials. Semiconductors are a small and shrinking fraction of it, and even the silicon aboard the machine is majority motor-control rather than AI compute. Chinese suppliers hold a decisive structural cost advantage in precisely the mechanical categories that dominate the machine, arising from end-to-end domestic supply chains and near-total control of rare-earth magnets. A Western strategy predicated on owning the intelligence layer is a strategy predicated on owning the smallest part of the robot.

The third is that the binding constraint is physical, not algorithmic. Power, memory and bandwidth - not model quality - determine what a robot can do away from a wall socket and a network connection. A full manipulation stack demands more memory than onboard accelerators can supply, and every watt spent on inference is a watt not spent on motion. The operational compute and memory layer that follows from this is the fastest-growing category in the components market, it has no incumbent, and it appears on no published market map.

The components layer is where the machine's cost actually lives, and it is the least funded part of the industry. That asymmetry is the central investment finding of this report.

The Global Emerging Robotics Market 2027–2037 is a comprehensive analysis of the seven verticals reshaping physical automation, the three-layer stack beneath them, and the capital being deployed against both. Published by Future Markets, Inc., the report combines a full quantitative forecast to 2037 with an unusually direct assessment of what these machines can and cannot presently do. The report introduces the Autonomy Dependency Scale (A0–A4), a classification applied consistently to every company profiled, distinguishing designed capability from observed operating class. It records, for the first time in a market study of this kind, that nothing in commercial deployment operates at A4, that the most autonomous systems on the market are in defence rather than humanoids, and that the gap between vendor claims and field performance is widest precisely where capital is most concentrated.

A detailed bill-of-materials analysis resolves the value-capture dispute that has dominated the sector's commentary. The report demonstrates that actuation accounts for 73% of a humanoid's cost, that semiconductors fall from 8% to 5%, and that cost share, margin and defensibility are three different things - with a 40–60% Chinese cost advantage in the categories that matter most. The forecast covers 2027–2037 across seven verticals, seven component categories, five business models and five regions, in base, conservative and optimistic scenarios, with a full sensitivity analysis identifying the consumer humanoid price threshold as the single largest variable in the market.

Contents:

• Executive summary - key findings, the three structural claims, and what changed since 2026
• Introduction and taxonomy - defining emerging robotics; the market map; the three-layer stack (components → platforms → intelligence)
• The autonomy gap - the Autonomy Dependency Scale A0–A4; designed versus observed autonomy by vertical; the data scarcity problem; why defence leads
• Value capture and the bill of materials - full BOM decomposition; the actuation share; the semiconductor share; cost share versus margin versus defensibility; the Chinese cost advantage quantified
• Supply chain and components - actuators and transmissions; dexterous hands (deep dive: $629M → $19.9bn, 41.3% CAGR); sensors; power systems; edge silicon; rare-earth and battery concentration; operational compute and memory
• Robotics foundation models - the merchant model market; the vertical-integration squeeze; the data problem; why the layer's share of value declines
• Humanoids - the three-wave adoption model; shipments versus ASP; the deployment reality check; concentration and the relocation of the market to China
• Warehouse and logistics - sub-segments; the pick-rate frontier; cost per pick; the RaaS transition; why the humanoid loses here
• Manufacturing and automation - the cell cost stack; batch-size economics; the skilled-trade shortage; the programming-cost curve
• Defence - attritable mass; designed versus observed autonomy in Ukraine; the power budget; counter-UAS economics; autonomy licensing
• Space - light-time delay and the collapse of teleoperation; the radiation-hardened compute gap; on-orbit servicing, ISAM and debris removal
• Construction and infrastructure - the fifty-year productivity divergence; the automation frontier; why solar is the wedge
• Market forecasts 2027–2037 - by vertical, component, business model, region and units; three scenarios; sensitivity analysis; the double-counting convention
• Investment and competitive landscape - VC trajectory; capital allocation versus revenue opportunity; the starved components layer; exit environment
• Company profiles - 168 companies with autonomy classification. Companies Profiled include 1X Technologies, ABB, Agibot, Agile Robots, Agility Robotics, AheadForm, AIRSKIN, AI² Robotics (AI2), AmbiRobotics, Anduril Industries, ANYbotics, Apptronik, ARX Robotics, Aubo Robotics, Augmentus, Baidu, BHRIC (Beijing Humanoid Robot Innovation Center), Boardwalk Robotics, Boost Robotics, Booster Robotics, Boston Dynamics, BridgeDP Robotics, Bright Machines, BRINC, Built Robotics, BXI Robotics, Charge Robotics, ClearPath Robotics, ClearSpace, Clone Robotics, Cognibotics, Contoro Robotics, Cosmic Robotics, Covariant, Daimon Robotics, Dataa Robotics, Deep Robotics, DeepCloud AI, Dexory, Dexterity, Diligent Robotics, Dobot Robotics, Doosan Robotics, dRobotics, Dusty Robotics, Dyna Robotics, Electron Robots, Elephant Robotics, EngineAI, Epoch Robotics, Eureka Robotics, EX Robots, Exotec, Fanuc, FBR (Hadrian X), FDROBOT, FESTO, Field AI, Figure AI, Fluid Wire Robotics, Formant, Forterra, ForwardX, Foundation, Fourier Intelligence, Franka Emika, Galaxea AI, Galbot, Gecko Robotics, Ghost Robotics, GITAI, GrayMatter Robotics, Hadrian, HavocAI, HEBI Robotics, Honda, Humanoid, Hypercraft, Icarus Robotics, Inivation, IntBot, intuiCell, Jacobi Robotics and more....

Table of Contents

1 EXECUTIVE SUMMARY

• 1.1 The Market in Summary
• 1.2 Principal Findings
  • 1.2.1 No General-Purpose Robot Exists, and None Is Close
  • 1.2.2 The Most Autonomous Systems on the Market Map Are in Defence, Not in Humanoids
  • 1.2.3 Is Value Concentrated in Actuation, and is the Concentration Structural
  • 1.2.4 The Binding Constraint Is Physical, Not Algorithmic
• 1.3 Market Forecast Summary
• 1.4 Implications for Positioning

2 INTRODUCTION: MARKET DEFINITION, TAXONOMY AND THE EMERGING ROBOTICS STACK

• 2.1 Defining Emerging Robotics
• 2.2 The Commercial Consequence of the Definition
  • 2.2.1 The Five Competing Business Models
• 2.3 The Seven-Vertical Taxonomy
  • 2.3.1 Foundation Models as a Layer Rather Than a Vertical
  • 2.3.2 Humanoids as a Form Factor Rather Than a Market
  • 2.3.3 The Growing Primacy of Defence
• 2.4 The Emerging Robotics Stack
  • 2.4.1 The Operational Compute and Memory Layer
  • 2.4.2 The Physical Components Layer
• 2.5 Scope Exclusions
• 2.6 Methodology and Basis of Estimates
  • 2.6.1 The Treatment of Pilots
  • 2.6.2 The Treatment of Teleoperated Systems
  • 2.6.3 The Treatment of Replacement Demand

3 THE AUTONOMY GAP: WHY NO GENERAL-PURPOSE ROBOT EXISTS

• 3.1 The Present State of General-Purpose Capability
• 3.2 The Four Senses of "Autonomous"
  • 3.2.1 The Consequence of Definitional Slippage
• 3.3 Teleoperation as Data Supply
  • 3.3.1 Teleoperation Within the Cost of Goods Sold
  • 3.3.2 The Economics of the Pilot
  • 3.3.3 The Step Function
• 3.4 The Autonomy Dependency Classification
  • 3.4.1 Application Across the Market Map
  • 3.4.2 The Primacy of Defence in Demonstrated Autonomy
  • 3.4.3 The Constraint Is Not Model Quality
• 3.5 The Infrastructure Dependency Problem
  • 3.5.1 Availability
  • 3.5.2 The Onboard Memory Wall
  • 3.5.3 Accountability and Operational Memory
• 3.6 The Counter-Argument
  • 3.6.1 Improving Connectivity
  • 3.6.2 The Trajectory of Onboard Silicon
  • 3.6.3 An Institutional Answer to Accountability
  • 3.6.4 Assessment
• 3.7 Consequences for the Forecast
  • 3.7.1 Humanoids
  • 3.7.2 Defence
  • 3.7.3 Foundation Models

4 VALUE CAPTURE: BILL-OF-MATERIALS ECONOMICS

• 4.1 The Dispute
• 4.2 What the Bill of Materials Actually Shows
  • 4.2.1 The Mechanical Dominance of Cost
  • 4.2.2 The Concentration Is Structural, Not Transitional
  • 4.2.3 Even the Silicon Is Mostly Actuation
• 4.3 The Error Common to Both Camps
  • 4.3.1 Cost Share Is Not Margin Capture
  • 4.3.2 Margin Capture Is Not Defensibility
• 4.4 Where the Defensible Positions Actually Lie
  • 4.4.1 High-Performance Actuation, Not Actuation
  • 4.4.2 Operational Compute and Memory
  • 4.4.3 Integration, Which Nobody Is Arguing For

5 THE SUPPLY CHAIN AND COMPONENTS LAYER

• 5.1 The Layer the Market Map Omits
• 5.2 Actuators, Harmonic Drives and Transmissions
• 5.3 End Effectors and Dexterous Hands
  • 5.3.1 The Dexterous Hand Market
  • 5.3.2 Why Hands Cost What They Cost
  • 5.3.3 Demand by Industry Application
  • 5.3.4 The Dexterity-Price Frontier
  • 5.3.5 Competitive Implications
• 5.4 Sensors and Perception
• 5.5 Power Systems and Operational Energy
  • 5.5.1 Power as Product
• 5.6 Semiconductors and Edge Compute
  • 5.6.1 The Memory Wall
  • 5.6.2 The Power Budget Trap
• 5.7 Operational Memory and Verifiable Autonomy
  • 5.7.1 The Requirement
  • 5.7.2 The Market Position
  • 5.7.3 The Bear Case
• 5.8 Geographic Concentration and Chokepoints

6 FOUNDATION MODELS AND ROBOT LEARNING

• 6.1 The Layer and Its Ambition
• 6.2 The Data Problem
  • 6.2.1 The Consequence for the Business Model
  • 6.2.2 Simulation as Partial Escape
• 6.3 The Vertical Integration Squeeze
• 6.4 The Inference Constraint
• 6.5 Competitive Landscape

7 HUMANOIDS

• 7.1 Market Overview
• 7.2 The Deployment Reality
• 7.3 The Three-Wave Structure
  • 7.3.1 Wave 1: Industrial
  • 7.3.2 Wave 2: Consumer and Developer
  • 7.3.3 Wave 3: Medical and Assistive
• 7.4 The Capability Gap
  • 7.4.1 The Manipulation Bottleneck
  • 7.4.2 The Step Function
• 7.5 Competitive Structure
  • 7.5.1 Why China Wins on the Current Cost Structure

8 WAREHOUSE AND LOGISTICS

• 8.1 Market Overview
• 8.2 The Task-Scripted Ceiling, and Why It Does Not Matter Here
• 8.3 The Economics: Cost Per Pick
• 8.4 What the Buyer Is Actually Buying
• 8.5 Business Model: The RaaS Transition
• 8.6 Competitive Landscape

9 MANUFACTURING AND AUTOMATION

• 9.1 Market Overview
• 9.2 The Real Cost of a Robot Is Not the Robot
• 9.3 The Batch-Size Window
• 9.4 The Labour Constraint Is Specific, Not General
• 9.5 Competitive Landscape

10 SPACE ROBOTICS

• 10.1 Market Overview
• 10.2 Why Space Cannot Cheat
• 10.3 The Radiation-Hardened Compute Gap
• 10.4 The Autonomy Reality
• 10.5 Competitive Landscape

11 CONSTRUCTION AND INFRASTRUCTURE

• 11.1 Market Overview
• 11.2 Why Construction Resisted
  • 11.2.1 The Site Is the Anti-Warehouse
  • 11.2.2 The Buyer Cannot Fund It
  • 11.2.3 The Labour Question Is Political
• 11.3 The Automation Frontier
• 11.4 Competitive Landscape

12 DEFENCE AND SECURITY

• 12.1 Market Overview
• 12.2 The Procurement Inversion
• 12.3 The Gap Between Designed and Observed Autonomy
• 12.4 The Domains
  • 12.4.1 Ground
  • 12.4.2 Air
  • 12.4.3 Maritime
  • 12.4.4 Counter-UAS
• 12.5 Company Landscape

13 COMPANY PROFILES (168 company profiles)

14 REFERENCES