Humanoid robotics is experiencing one of those moments where technology seems to be advancing faster than the companies called to adopt it. Prototypes walk, carry boxes, recognize objects, receive voice commands, and begin to combine artificial intelligence models with sensors, actuators, and imitation learning. But between a viral demo and a profitable deployment in a factory, warehouse, or hospital, there’s still a considerable gap.
McKinsey has named this bottleneck: “pilot purgatory.” The consulting firm warns that humanoid and general-purpose robots are no longer just a technical curiosity, but many organizations still lack clear processes, infrastructure, talent, or economic criteria to bring them to production. The industry looks at a market that could approach $400 billion by 2040, while companies continue wondering what to do with robots that impress in laboratories but don’t always justify their costs in real operations.
The message was strongly emphasized at the 5th Mobis Mobility Day, held in Sunnyvale, Silicon Valley, focusing on physical robotics and artificial intelligence. Hyundai Mobis brought together startups, investors, manufacturers, and experts in an event that reflects the sector’s direction: robotics is no longer just industrial machinery, but a physical extension of AI.
Physical AI is no longer science fiction, but it’s also not plug and play
The key to this new wave is embodied AI. Unlike generative AI that writes texts, creates images, or summarizes documents, physical AI must act in the real world. It needs to interpret cameras, touch sensors, force, position, balance, obstacles, human commands, and unpredictable environmental changes. And then, move a mechanical body without breaking anything, harming anyone, or stopping a production line.
McKinsey notes that general-purpose robots are gaining attention due to improvements in vision-language-action models, multimodal sensors, and the ability to learn by observing humans. In theory, this allows a robot to interpret an instruction, recognize an object, and perform a task without step-by-step programming for each movement. In practice, errors still occur when the environment changes, objects aren’t where they should be, or tasks require fine manipulation.
Investor enthusiasm is genuine. McKinsey estimates that annual funding for general-purpose robotics has surpassed $1 billion since 2022, and patent filings have grown at a compounded annual rate of 40%. The potential market is also projected around $370 billion by 2040, with China as one of the major growth hubs. It’s a huge figure but contingent on technology crossing several still-open thresholds.
| Key Data on General-Purpose Robotics | Approximate Figures |
|---|---|
| Estimated potential market by McKinsey in 2040 | $370 billion |
| Annual funding since 2022 | Over $1 billion |
| Patents growth rate since 2022 | 40% |
| Typical current cost of humanoid prototypes, per McKinsey | $150,000 to $500,000 |
| Target cost to compete in mass sectors | $20,000 to $50,000 |
| Typical autonomy of many current humanoids | 2 to 4 hours |
| Actuation’s share of total robot cost | 40% to 60% |
The problem is that robots don’t arrive at a company the way a SaaS application does. It’s not enough to buy a license, train a team, and connect data. A robot requires physical safety, maintenance, spare parts, operation zones, integration with internal systems, emergency protocols, human supervision, workforce acceptance, and return on investment calculations. If it fails, it’s not just an error on a screen: it could hit a part, stop a line, or cause an accident.
Four barriers before leaving the lab
McKinsey summarizes the leap to widespread deployment into four major barriers: safety, autonomy, dexterity, and cost. None are minor.
Safety is the first. Humanoid robots promise to operate in spaces designed for humans, without fences or complete reengineering of warehouses, hospitals, or factories. That’s their main advantage over traditional industrial robots but also their greatest risk. A humanoid moving among workers must detect human presence, avoid collisions, limit exerted force, react to falls, and operate under clear standards. Current collaborative robotics standards do not cover all cases for autonomous humanoid robots in open environments.
Autonomy is the second barrier. Many current humanoids do not operate a full industrial shift. If a robot works two or four hours and then needs long charging times, its real productivity drops. Solutions include swappable batteries, fast charging, more efficient designs, and planned cycle operations, but a mature solution for long shifts is still lacking.
The third barrier is dexterity. Moving around a warehouse can be complex; manipulating irregular objects is much harder. The human hand combines strength, sensitivity, speed, and precision, which are difficult to replicate. A robot can lift boxes or carry parts, but tying shoelaces, peeling fruit, handling fragile objects, or working with small components in awkward positions remains very challenging. Practically, initial use cases should focus on repetitive, structured, low-variability tasks.
The fourth barrier is cost. Current prototypes are expensive due to immature components, limited supply chains, and still-low industrialization. McKinsey estimates actuation—motors, gearboxes, joints, sensors involved in movement—to account for 40% to 60% of total costs. To make robots economically viable outside specific pilots, prices must fall sharply, and maintenance needs to become predictable.
The risk of perpetual testing
“Pilot purgatory” isn’t exclusive to robotics. It’s also happened with generative AI, IoT, augmented reality, and advanced automation. A company tests a technology, gets a promising demo, presents it internally, and then struggles to scale because of lacking budget, clear responsibilities, integration, or business metrics.
In robotics, this risk is even greater because deployment involves physical operations. It’s not enough to show the robot can perform a task. It’s essential to know how long it takes, failure rates, repair responsibilities, consequences of falls, coordination with humans, shutdown costs, necessary insurance, data collection, and real productivity gains compared to human or traditional automation alternatives.
Therefore, the most sensible recommendation is not to buy robots just because it’s fashionable but to choose narrow, measurable use cases. A warehouse could start with internal transport of light loads along defined routes. A factory might try feeding machines or moving parts in controlled zones. A hospital can evaluate logistics tasks that don’t involve direct patient contact. The key is avoiding overly ambitious projects that turn robots into expensive, useless attractions.
Humanoid robotics also needs a more mature supply chain. Critical components like precision actuators, tactile sensors, gearboxes, or high-load planetary screws don’t always scale with projected demand. China has an advantage here, as was the case with batteries, electric vehicles, and consumer electronics. For Europe and the US, the challenge is not only developing good robots but also ensuring suppliers, standards, and manufacturing capacity.
Hyundai Mobis’ interest in robotics and physical AI fits into this context. The company is seeking technological alliances and investment opportunities in a field potentially crucial for mobility, manufacturing, components, industrial automation, and services. The presence of startups working on humanoid arms, perception, and AI models at their Silicon Valley event shows that automotive, robotics, and AI are beginning to share a common agenda.
The promise of humanoid robotics is compelling: machines capable of working in human spaces, learning new tasks, and assisting in sectors with labor shortages, physical risks, or low productivity. But the market will grow only when robots are safe, can operate full shifts, manipulate objects reliably, cost less, and integrate into real operations.
The opportunity exists but won’t happen automatically. Companies aiming to generate value will need to prepare processes, data, personnel, maintenance, and ROI criteria before the technology is fully mature. Humanoid robotics could become one of the major industries of the coming decades or remain in a series of brilliant pilots that never go beyond prototypes.
Frequently Asked Questions
What does “pilot purgatory” mean in robotics?
It refers to the situation where a company tests robots in pilot projects or demos but cannot scale them to full production due to economic, safety, integration, or operational readiness issues.
How much could the general-purpose robotics market be worth?
McKinsey estimates it could reach around $370 billion by 2040 if the technology continues evolving and overcomes barriers related to cost, safety, autonomy, and adoption.
Why aren’t humanoid robots yet widely deployed?
Because they still face limitations in autonomy, dexterity, safety in open spaces, manufacturing costs, maintenance, integration with existing systems, and organizational acceptance.
Which sectors might adopt humanoid robots first?
Initial deployments are likely in structured environments such as logistics, manufacturing, inspection, simple manipulation, repetitive tasks, and operations with labor shortages or physical risks.