Two bipedal robots are about to step onto one of the busiest tarmacs in Asia. Starting in May 2026, Japan Airlines and its ground-handling subsidiary JAL Ground Service will station a pair of Unitree G1 humanoid machines at Tokyo’s Haneda Airport, tasking them with loading and unloading checked baggage and cargo alongside human ramp crews. The trial, organized with technology partner GMO AI and Robotics, is described by the three partners as the first time humanoid robots have been used in airport ground operations anywhere in Japan, though no independent source has confirmed that claim.
The project lands at a moment when airlines worldwide are struggling to recruit and retain ramp agents, workers who hoist suitcases weighing up to 23 kilograms each, often in extreme heat or cold, across shifts that commonly stretch beyond eight hours. At Haneda alone, which handled roughly 87 million passengers in fiscal year 2024, the physical toll of that work has made turnover a persistent problem. Japan’s shrinking working-age population, projected by the National Institute of Population and Social Security Research to decline by about 12 million people between 2020 and 2040, only sharpens the pressure.
What the partners have confirmed
A joint press release issued in late April 2026 describes the Haneda experiment as a demonstration-phase project. The two G1 units will initially handle defined tasks: transferring individual suitcases between baggage carts and conveyor belts and assisting with loading in the lower-deck holds of narrow-body aircraft. Human supervisors will monitor every movement and can halt operations immediately if a safety concern arises.
Future phases could expand into cabin cleaning and the operation of ground support equipment such as belt loaders and pushback tractors, though no firm dates have been set for those additions. The press release frames the effort as an initial experiment with aspirations for broader application; it does not confirm a fixed multi-year timeline. Early coverage citing a “three-year rollout” appears to reflect the partners’ stated ambitions rather than a contractual commitment.
The robots themselves come from Unitree Robotics, a Chinese manufacturer whose product announcement lists the G1 at roughly $16,000 before tax and shipping. Adjusted for yen-to-dollar conversion at recent exchange rates, that figure aligns with the approximately $15,400 per unit cited in early reporting. GMO AI and Robotics, a division of the GMO Internet Group that recently opened the GMO Humanoid Lab in Shibuya to showcase enterprise robotics, is responsible for supplying the hardware and writing the motion programs that will guide the machines through ramp-specific tasks.
The battery problem and other open questions
Unitree rates the G1’s battery life at about two hours per charge. That number comes from the manufacturer’s own spec sheet and has not been independently verified under the kind of load an airport ramp would impose. Repeatedly lifting 23-kilogram bags in summer heat or winter wind would almost certainly drain a battery faster than a climate-controlled lab demo. Whether JAL plans to rotate robots through charging stations, test extended-capacity battery packs, or simply limit the machines to short task bursts has not been disclosed.
Other gaps in the public record are just as significant:
- Staffing impact: The announcement highlights labor savings and efficiency gains but does not quantify expected headcount reductions or describe how human crews will share space with the robots during live aircraft turnarounds.
- Total cost of ownership: The $16,000 list price covers hardware only. Enterprise robotics deployments typically add software licensing, custom programming, sensor calibration, spare parts, and on-site technical support. GMO’s role in motion-program development signals that substantial customization sits on top of the sticker price.
- Regulatory clearance: Neither the press release nor any public statement from Japan’s Civil Aviation Bureau (JCAB) indicates whether specific guidance has been issued for humanoid robots operating in restricted airport zones, or how liability would be allocated if a machine damaged an aircraft or piece of ground equipment.
- Wireless and sensor coexistence: Active ramps are dense radio environments. How the G1’s onboard sensors and communications will interact with existing airport systems designed around human operators and conventional vehicles remains unaddressed.
Direct statements from ramp workers, union representatives, JAL operations managers, robotics engineers, or independent labor analysts have not appeared in the available record. The absence of any on-the-ground human voices is a notable gap; without those perspectives, the real-world effect on daily workflows at Haneda is difficult to gauge.
How Haneda fits into a wider automation push
JAL’s trial is not happening in isolation. Airports and logistics companies have been testing robotic systems for ground-side work for several years. DHL has piloted Boston Dynamics’ Stretch robot for container unloading at distribution centers. Singapore’s Changi Airport uses automated guided vehicles to move baggage between terminals. Vanderlande and other baggage-system manufacturers have steadily increased the automation of sorting and screening lines.
What distinguishes the Haneda project is the form factor. Most existing airport automation relies on wheeled platforms, conveyor extensions, or robotic arms bolted to fixed stations. A humanoid that walks on two legs and manipulates bags with articulated hands is, at least in theory, designed to slot into spaces built for human bodies: the cramped belly of a Boeing 737, the narrow gap between a baggage cart and a belt loader. Whether that theoretical advantage holds up against the irregular geometry of real cargo holds, where bag sizes vary and containers shift mid-load, is precisely what the trial is meant to test.
The economics also differ from most warehouse robotics. At roughly $15,400 per unit for hardware alone, the G1 is far cheaper than industrial robotic arms or autonomous mobile platforms that can cost six figures. If the total deployment cost, including software, maintenance, and supervision, stays low enough to undercut the fully loaded cost of a ramp agent’s wages and benefits, the business case could attract attention from ground handlers beyond Japan. If integration and upkeep push the real price well above the headline number, the calculus changes quickly.
What the trial still needs to prove
The most useful data from Haneda will not come from press releases. It will come from granular operational logs: bags moved per hour compared to a human handler, frequency of manual interventions, battery degradation across repeated outdoor charge cycles, and any safety incidents or near-misses on the ramp. None of that data exists yet. The experiment is scheduled to begin in May 2026, and meaningful performance trends will take months to accumulate.
For airlines and airport operators watching from other hubs, two robots on one ramp at one airport will not reshape the global ground-handling workforce. But credible efficiency numbers, published transparently and including failure modes, could accelerate similar pilots at airports facing the same staffing crunch. The opposite outcome is equally plausible: if the G1 units prove too slow, too fragile, or too complex to weave into existing turnaround choreography, the trial could reinforce skepticism about humanoid machines in heavy-duty industrial settings.
How much the rest of the industry learns depends largely on how JAL and GMO AI and Robotics choose to share results. Detailed reporting on uptime, breakdowns, and worker feedback would give ground handlers everywhere a realistic picture of what humanoid robots can and cannot do on the ramp. Selective case studies would leave everyone guessing. Until that evidence surfaces, the Haneda project is best understood as an ambitious, closely watched experiment, not a settled blueprint for the future of airport work.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
