Unitree G1 at Haneda Airport: Chinese humanoid robots make their way to Japan — physical robotics scales up

Unitree G1 at Haneda Airport: Chinese humanoid robots arrive in Japan — physical robotics scales up

🔎 Japan imports its robots: a major geopolitical symbol

In May 2026, Japan Airlines deployed Chinese humanoid robots on the tarmac of its largest airport. Not in a lab. Not in a demo. On the Haneda tarmac, next to Boeing 787s, with real passengers' luggage.

This is a symbolic earthquake. Japan — the country that invented Asimo, the historical champion of robotics — is calling on Unitree, a startup from Hangzhou, to solve a problem that its own giants (Fanuc, Yaskawa, Kawasaki) failed to address in time. Economic pragmatism prevailed over technological nationalism.

The pilot runs for two years (May 2026 – May 2028). Two models are deployed: the Unitree G1 for baggage handling and the UBTECH Walker E for cabin cleaning. The objective is clear: test whether teleoperated humanoids can absorb part of the physical ground work in a constrained airport environment.

The essentials

• Japan Airlines deploys Unitree G1 and UBTECH Walker E at Tokyo Haneda Airport as part of a 2-year pilot that started in May 2026.
• The Unitree G1 (132 cm, 35 kg, aerospace aluminum + carbon) is used for sorting and transporting luggage, including ULD containers. Cost per unit: approximately $15,400 according to Kraneshares.
• The robots are teleoperated by human controllers — full autonomy is not yet achieved, but human-robot coordination on the tarmac is an operational first.
• Japan chose China out of pragmatism: immediate availability, competitive pricing, and the inability of Japanese manufacturers to offer a humanoid equivalent ready to deploy.
• The humanoid robot market is estimated at over $50 billion by 2030 (Goldman Sachs projects $38 billion by 2035, Morgan Stanley $152 billion by 2040 according to AI Magicx).

Recommended tools

| Unitree G1 | Baggage handling, airport tarmac | ~$15,400 (May 2026, check on unitree.com) | Industrial deployments at a reduced price |
| UBTECH Walker E | Cabin cleaning, heavy ground work | Price not advertised (May 2026) | Physical tasks in constrained environments |
| Tesla Optimus Gen 3 | Factory, logistics | ~$25,000 (May 2026, check on tesla.com) | Tesla ecosystem integration, production scaling |
| Boston Dynamics Atlas | Factory, heavy handling | Commercial lease (May 2026, check on bostondynamics.com) | Complex tasks, industrial environment |

The Haneda deployment: what is actually happening on the tarmac

The Unitree G1 pushes luggage containers on a conveyor belt next to a JAL plane. It loads ULD (Unit Load Device) containers, those standardized crates that are slid into the cargo hold of commercial aircraft.

The scene is documented by Interesting Engineering: a child-sized robot (132 cm) handles air cargo in an environment where the slightest error can delay a flight. This is not a cleanroom demonstration. It is an operational test with real constraints — noise, weather, safety.

The G1 weighs 35 kg, built with aerospace aluminum and carbon fiber according to Gadget Review. It moves at 7 km/h and operates for about 2 hours per charge. Limited autonomy, but sufficient for defined work cycles on the tarmac.

UBTECH's Walker E, larger (adult size), is assigned to cabin cleaning — a physically grueling task that airlines struggle to staff. Yahoo notes, however, that the robots "don't seem quite ready yet," highlighting visible hesitations in certain movements.

Teleoperation: not yet an autonomous robot

A crucial point often misunderstood: the Unitree G1s at Haneda are not autonomous. They are teleoperated by remote human controllers. The robot executes the physical movements, but a human decides on every action.

This is a smart transition model. JAL is not asking the robot to "understand" the airport environment — an immense AI challenge. It is asking the robot to be the physical arm of an operator who fully understands the context. It is the same logic as robotic surgery: the Da Vinci does not decide, it executes with superior mechanical precision.

This teleoperated model could even become the industrial standard before full autonomy. Kraneshares analyzes this dynamic as the transition "from pilot to platform": humanoids start as teleoperated tools, then gradually gain autonomy thanks to the accumulation of operational data.

Why Japan chose China: JAL's cold calculation

Japan is the country of Fanuc, Yaskawa, Kawasaki Heavy Industries — global giants of industrial robotics. Yet, JAL turned to Unitree (Hangzhou, China) and UBTECH (Shenzhen, China). The economic calculation is unemotional.

Price as a geopolitical weapon

The Unitree G1 costs around $11,700 to $15,400 per unit depending on configurations and sources (RoboZaps cites ~$13,500, Kraneshares cites ~$15,400). At this price, JAL can deploy several of them without an astronomical investment.

Japanese robotics giants do not have a humanoid equivalent at this price point. Fanuc dominates industrial robotic arms (the cobot and articulator market), but has no commercialized bipedal humanoid for handling. Yaskawa is in the same boat. The Chinese product filled a Japanese void.

Immediate availability

China shipped over 10,000 humanoid robots in 2025 according to Top Tech News. The Chinese ecosystem has reached a production capacity that no one else has matched yet. When JAL needs 10 robots for a pilot at Haneda, Unitree can deliver. Whereas a Japanese manufacturer would first have to develop the product.

Pragmatism prevailed over nationalism. This is a strong signal: in physical robotics, just like with solar panels or batteries, China has taken a manufacturing lead that Japan will not catch up on in the humanoid segment.

The labor shortage: an urgency that leaves no choice

Japan is aging. Its working population is shrinking. Tourism is exploding — the country has broken its international visitor records. Ars Technica summarizes the situation: Japan is turning to China to cope with the labor shortage. Luggage doesn't sort itself, and JAL can no longer find enough ground workers.

Unitree G1 vs Tesla Optimus vs Boston Dynamics Atlas : the humanoid race in 2026

The Haneda deployment is not an isolated case. It is part of a pivotal year where several humanoids transition from prototype to real-world deployment. BlueHeadline is categorical: "Real metal is walking in real factories."

Comparison of the main humanoids deployed in 2026

Tesla Optimus : the vertical integration approach

Tesla unveiled the Optimus Gen 3 in Q1 2026 according to Top Tech News. At ~$25,000 according to RoboZaps, it costs almost twice as much as the G1. But Tesla is betting on the ecosystem: the same batteries as its cars, the same manufacturing infrastructure, the same software stack.

Tesla's advantage is volume. If Optimus reaches its targeted mass production, the unit cost could drop below that of the G1. But in May 2026, Unitree already has robots in service. Tesla is still in its internal scaling phase.

Boston Dynamics Atlas : the robotics premium

Qviro highlights the remaining challenges for Atlas, despite its entry into production in 2026. AI Magicx confirms that Boston Dynamics now offers commercial leasing for the electric Atlas — a "Robotics-as-a-Service" model that avoids the upfront purchase.

Atlas remains the most technically capable in terms of mobility and dexterity. But its cost (lease + maintenance) reserves it for high-value-added applications. JAL could not have deployed 10 Atlas units at Haneda for a pilot — the budget wouldn't have allowed it.

Unitree G1 : the price-availability play

This is where the G1 wins. It is not the most capable. It is not the fastest or the most autonomous. But it is available now, at a price that enables multi-unit deployment, and it does the job asked of it — pushing luggage on a conveyor belt. This dynamic echoes the history of the Chinese auto industry: initially underpriced and criticized, then gradually improved until becoming competitive on the grounds of quality.

The humanoid robot race: where do we stand in 2026?

2026 is the year humanoid robotics shifts from laboratory demonstrations to operational deployment. Not everywhere. Not for all tasks. But enough to make the narrative shift irreversible.

Top Tech News identifies 2026 as the "tipping point" for humanoid robots. Three signals converge: China has industrialized production, the United States has industrialized onboard AI, and client companies (JAL, BMW, Tesla) are willing to pay for real-world pilots.

Concrete deployments in May 2026

• Haneda (Tokyo): Unitree G1 + UBTECH Walker E for JAL — baggage handling and cabin cleaning.
• Tesla factories: Optimus Gen 3 deployed internally for logistics and assembly tasks.
• BMW factories: Figure 02 operates on assembly lines according to BlueHeadline.
• Unspecified factories: Boston Dynamics Atlas in commercial production via leasing.

The diversity of deployments is significant. We are no longer talking about a single robot in a single context. We are talking about multiple robotic platforms in different industrial environments — airport, automotive, logistics.

What this means for Figure 02 et la course aux robots humanoïdes : qui gagne ?

The Haneda deployment adds a player (Unitree) to a race often presented as a Tesla vs Boston Dynamics duel. In reality, the market is segmenting. The high end (Atlas, Figure 02) targets full autonomy in complex environments. The mid-range (Optimus) aims for internalized mass production. The low end (G1) bets on price and availability for teleoperation.

Each segment has its use case. An airport doesn't need a robot capable of solving an unprecedented problem — it needs a reliable robot that pushes luggage 8 hours a day. The G1 meets these specifications.

Onboard AI: the link with agentic models

The transition from teleoperation to autonomy passes through onboard AI. Current agentic models — GPT-5.5 (agentic score: 98.2), Gemini 3 Pro Deep Think (95.4), Claude Opus 4.7 Adaptive (94.3) — are not yet integrated into physical robots in a standardized way. But the trajectory is clear.

When a model like GPT-5.5 can plan a complex sequence of actions with a score of 98.2, the question is no longer "can AI think for a robot?" but "when will AI think fast enough and reliably enough to do so in real time on a physical robot?". The industry estimates this shift to 2027-2028 for structured industrial tasks.

Figure AI lance Helix : le système d'IA qui amène les robots humanoïdes dans les foyers perfectly illustrates this convergence: an AI system designed specifically to pilot a robotic body, not just generate text.

Geopolitical stakes: when robotics becomes a matter of sovereignty

JAL's choice is not just a purchasing decision. It is a geopolitical signal. Japan, a strategic ally of the United States, is importing Chinese robots for its critical airport infrastructure.

The robotic supply chain

The components of the Unitree G1 — actuators, sensors, compute — are predominantly Chinese. This means JAL depends on a supply chain over which Tokyo has no direct regulatory control. In the event of geopolitical tension (Taiwan, tech sanctions), the supply could be interrupted.

This is the paradox: Japan imports Chinese robots precisely because it cannot manufacture enough of them itself, creating a dependency it sought to avoid.

Japan's likely response

Fanuc and Yaskawa will not sit idle. The Haneda deployment is an electric shock for the Japanese robotics industry, which has dominated the industrial arm market but missed the humanoid turn. Genesis AI dévoile GENE-26.5 et des mains robotiques humanoïdes : la robotique passe full-stack shows that innovation in humanoid robotics is not monopolized by a single country — but low-cost production capacity remains a decisive Chinese advantage.

Japan could respond with public subsidies, partnerships with emerging Japanese startups, or joint ventures with American manufacturers (Boston Dynamics, now under Hyundai's control). But the catch-up time is estimated at a minimum of 2-3 years.

Where do the United States fit in?

Washington is watching. If Chinese humanoid robots end up in the airports, factories, and warehouses of American allies, technological dependency extends beyond semiconductors and telecoms. US export controls on AI chips (Nvidia H100, etc.) partially slow down Chinese onboard AI, but the G1's mechanical hardware is not affected by these restrictions.

What the Haneda pilot teaches us about the future of physical work

Beyond geopolitics and technology, there are humans. JAL's ground workers at Haneda are not (yet) replaced — they are assisted by robots. But the direction of the pilot is unambiguous.

Jobs directly impacted

Ramp agents load luggage, push ULD containers, and clean cabins. These are physically demanding jobs, often outdoors, with irregular hours. In Japan, these positions are becoming increasingly hard to fill.

The Unitree G1 does not do all the work of a ramp agent. It cannot yet drive a baggage tractor, communicate with the crew, or react to a complex unforeseen event. But for the repetitive task of pushing containers on a conveyor belt, it is sufficient.

The most likely scenario: complementarity before substitution

In the short term (2026-2028, duration of the pilot), the model will be human + robot. An operator teleoperates the G1 from an air-conditioned control station — a less physical job, but one that requires new skills (controller, cameras, latency).

In the medium term (2028-2030), if autonomy progresses, a human will supervise several robots simultaneously. The ratio shifts from 1:1 to 1:3 or 1:5. This is where the impact on employment becomes real — fewer humans needed for the same volume of work.

In the long term (post-2030), in a scenario of full autonomy, the need for ground labor could decrease significantly. But this scenario assumes major advances in embedded AI that are not guaranteed.

Reskilling as the central issue

The humanoid robot teleoperator is a profession that did not exist five years ago. It could become a common job in the next decade. The question for JAL — and for all employers deploying humanoids — is to reskill its physical workers into teleoperators, rather than laying them off.

This is a political and managerial choice, not a technological one. Technology enables both scenarios.

The humanoid market: the numbers that matter

Financial projections give an idea of the scale of the transformation.

Goldman Sachs estimates the humanoid robot market at $38 billion by 2035. Morgan Stanley projects $152 billion by 2040. These figures, cited by AI Magicx, reflect different assumptions about the pace of adoption — but all converge toward a massive market.

Accelerating factors

• Falling unit cost: The G1 at ~$15,000 broke a psychological ceiling. If Tesla Optimus truly reaches $25,000 in mass production, the ROI becomes calculable for a growing number of companies.
• The global labor shortage: Japan is at the forefront, but Germany, South Korea, and even the United States are facing physical worker shortages in logistics and manufacturing.
• Improvement in embedded AI: LLM agentic scores show rapid progression (GPT-5.5 at 98.2 in 2025). Hardware-software integration is following suit.
• Field data: Each pilot like Haneda generates training data that improves subsequent models. It is a virtuous circle.

Inhibiting factors

• Reliability in unstructured environments: An airport is relatively structured. A construction site, a messy warehouse, a home — that's something else entirely.
• Regulation: No clear legal framework exists for the liability of a humanoid robot that injures someone or damages an aircraft.
• Social acceptance: Yahoo notes that passengers and workers observe the robots with skepticism. Trust is built slowly.

❌ Common mistakes

Mistake 1: Confusing teleoperation and autonomy

Many comments present the Haneda Unitree G1s as "autonomous robots." They are not. A human controls them remotely. Confusing the two completely skews the analysis of the technological level achieved. The solution: always check whether the robot is teleoperated, semi-autonomous, or fully autonomous before drawing conclusions.

Mistake 2: Dismissing the Chinese choice as anecdotal

"JAL just took the cheapest option" poorly summarizes the situation. Japan is the world's second-largest robotics market. For its flagship airport to use Chinese humanoids is a structural signal about Japan's loss of competitiveness in this specific segment. The solution: analyze the sourcing choice as a geopolitical indicator, not just a B2B purchase.

Mistake 3: Projecting full autonomy in the short term

Seeing a G1 push luggage and deducing that robots will replace all physical workers within 3 years is a projection error. The JAL pilot runs for 2 years in teleoperation. Autonomy for non-repetitive tasks in variable environments is estimated at 2028-2030 minimum. The solution: distinguish what exists today (teleoperation, repetitive tasks) from what is projected (generalized autonomy).

Mistake 4: Comparing the G1 to Atlas as if they were in the same category

The Unitree G1 (~$15,000) and the Boston Dynamics Atlas (premium rental) are not playing in the same league. Comparing their technical performance without adjusting for price and availability is misleading. The solution: always contextualize comparisons by the price/performance/availability ratio.

❓ Frequently asked questions

How many Unitree G1 robots are deployed at Haneda?

The exact number is not publicized by JAL. Sources mention a deployment involving several units of G1 and Walker E, but the precise figure was not communicated during the pilot announcement in May 2026.

Can the Unitree G1 work in the rain?

Sources do not specify the IP certification of the G1 deployed at Haneda. The tarmac being an outdoor environment, JAL has likely defined weather conditions for operation, but the technical details are not public at this stage.

Why didn't JAL choose a Japanese robot?

Because no Japanese manufacturer (Fanuc, Yaskawa, Kawasaki) offers a commercialized bipedal humanoid robot at a price and availability comparable to the G1. These companies dominate industrial robotic arms, not handling humanoids.

Can a robot really damage an aircraft?

This is one of the risks of the pilot. A poorly handled ULD container can damage an aircraft's cargo hold or fuselage. This is precisely why human teleoperation is used: the operator retains situational judgment. Liability in the event of damage is an issue that the 2-year pilot will need to clarify.

When will humanoid robots be autonomous in airports?

Estimates vary between 2028 and 2032 for reliable autonomy in an airport environment. The Haneda pilot is generating data that will accelerate this transition, but safety dictates a gradual approach.

✅ Conclusion

The deployment of Unitree G1s at Haneda is not an anecdote — it is the first operational chapter of humanoid robotics on an industrial scale. Japan made a pragmatic choice that disrupts its own technological dogmas, and the rest of the world is watching. If the two-year pilot confirms the model, every airport, every factory, every warehouse on the planet will have the same conversation in 2027. The question is no longer whether humanoid robots will arrive in our workplaces, but at what price and under which flag.