When Does a Humanoid Robot Make Business Sense?

China is putting embodied AI into factories, logistics and public spaces. The commercial test is whether a human-shaped machine can do enough different jobs to justify the extra cost and complexity.

China already produces most of the world’s humanoid robots. Nearly 90 percent of the more than 13’000 two-armed, two-legged robots delivered globally last year were made there, while the country has made embodied AI – artificial intelligence operating through physical machines — part of its industrial strategy.

The unresolved issue is not whether robots will perform more physical work. Companies already use specialised machines to move loads, inspect infrastructure and automate production. We think it is interesting to explore whether businesses gain enough from a robot that walks, reaches and handles objects like a person to compensate for the engineering required to make it do so.

Human-shaped robots have one clear advantage

Factories, warehouses, shops and offices were built around people. Stairs, doors, shelves, tools and workstations assume human height, reach and movement. A humanoid robot can enter those spaces without forcing a company to rebuild them around a machine.

Flexibility strengthens the case. A fixed robotic arm performs one defined sequence well. A humanoid that can carry a box, open a door and operate equipment could move between jobs instead of remaining attached to one production step.

Chinese companies are already testing both approaches. The Shanghai robotics centre covered by MIT Technology Review houses more than 100 companies developing machines for jobs ranging from transporting heavy loads to inspecting sewage pipes. Robotic arms at the same centre attempt tasks such as folding clothes, while humanoids make coffee and perform demonstrations for visitors.

A company that regularly changes products, layouts or tasks has more reason to pay for flexibility than a manufacturer repeating the same movement thousands of times.

Specialised machines keep an economic advantage

Two legs create problems as well as possibilities. Humanoids have to balance their weight, avoid people and objects, manage battery consumption and recover safely when something goes wrong. MIT Technology Review points to high prices, short battery life and safety problems as constraints, while some roboticists question whether the human form is the right design for many jobs. MIT Technology Review 09.10 202…

A warehouse moving identical pallets does not necessarily need legs. Wheels require less energy and provide greater stability. A factory repeating the same welding operation does not need a general-purpose pair of hands if a fixed arm performs the job faster.

Specialised robots also remove capabilities the buyer does not need. Fewer joints, sensors and moving parts can reduce purchase cost, maintenance and failure points. Humanoid economics improve only once flexibility replaces enough separate equipment or labour to offset those disadvantages.

Labour shortages change the calculation

Physical AI becomes more attractive where companies struggle to staff repetitive, dangerous or undesirable work. An inspection robot does not need a human shape if its only job is entering a sewage pipe. A machine working across several parts of a warehouse faces a different requirement. It has to move through human-designed spaces, identify changing objects and complete tasks that were previously assigned to different workers.

The same robot becomes more valuable if software allows a company to add tasks without buying another piece of hardware. This separates embodied AI from conventional industrial automation. The company is no longer paying only for one mechanical movement. It is paying for a physical platform that can learn additional work. Whether the economics hold depends on how often companies can actually reuse the same machine.

Demonstrations do not answer the procurement question

China’s robotics industry is making humanoids highly visible. Machines perform martial arts, run races and entertain crowds in shopping centres and tourist attractions. The demonstrations expose consumers and companies to the technology, but they provide little evidence about industrial productivity. MIT Technology Review 09.10 202…

A buyer needs operating data instead. How many hours does the robot work between charges? How often does an employee intervene? How long does it take to teach the machine another task? What happens after a fall or failed grasp? How much production stops while technicians repair it?

A robot that completes a controlled demonstration does not face the same conditions as one handling thousands of objects during a factory shift.

One number will decide many purchases

Cost per completed task gives companies a cleaner comparison than robot versus employee. The calculation includes the machine, software, integration, energy, maintenance, supervision and downtime. Businesses can then compare the result with a worker, a fixed robotic arm, a wheeled machine or another existing process.

A humanoid earns the extra engineering if one machine completes several jobs across an environment designed for people. If the work stays predictable and stationary, specialised automation has fewer reasons to give up its cost and reliability advantage.

China’s next phase of embodied AI will make the difference easier to see. Watch whether manufacturers move humanoids from demonstrations into paid industrial deployments, how many hours they operate without human intervention and whether companies start using the same robot for several tasks. Those measures will show whether human-shaped robots are becoming equipment businesses can justify rather than machines they want to watch.

BEFORE YOU BUY A HUMANOID ROBOT

Ask four questions first:

• Does the job require a human-shaped machine? If wheels or a fixed arm can do it, the extra engineering may not pay.
• How many different tasks will one robot perform? Humanoids make more sense when the same machine can move between jobs.
• How much human intervention does it need? Include supervision, charging, maintenance and recovery from errors in the cost.
• What is the cost per completed task? Compare the full operating cost with labour and existing automation, not the purchase price alone.