Robots at airports and stations: what 2026 trials in Tokyo, Osaka and Hong Kong showed
Three transport operators tested humanoid robots in 2026: baggage work at Haneda, passenger guidance in Osaka's metro and patrols in Hong Kong. What they reported, and what is still unknown.
Tác giả: Ban biên tập humanoid4rent · đã đăng 16 tháng 11, 2026
Three transport operators put humanoid robots into documented trials in 2026. Japan Airlines began a two-year ground-handling trial at Tokyo Haneda in May, Osaka Metro ran a two-week humanoid guide at two stations in September, and Hong Kong International Airport tested patrol humanoids until one fell down a staircase in August. None has yet published measured results, but each says something about how robots cope with crowded, safety-critical spaces.
This article sets out what each operator and its partners have stated, where reports disagree, and what remains unknown. It concentrates on transport spaces; our earlier piece on service robot pilots in European airports, shops and hotels covers retail and hospitality.
Haneda: humanoids behind the scenes at Japan Airlines
In late April 2026, Japan Airlines (JAL) announced that it would test humanoid robots in ground handling at Haneda, starting in May 2026 and running until 2028, according to heise online's report on the baggage trial. The planned tasks are loading baggage onto conveyor belts, cleaning aircraft cabins and operating ground support equipment. At the launch JAL showed a robot pushing cargo onto a belt next to an aircraft.
JAL's stated reasoning is practical. Ground handling is heavy physical work, Japan has a growing labour shortage, and a robot with a human-like range of motion can in principle work around existing aircraft and airport equipment without changes to the infrastructure.
The robot model is not reported consistently. heise's baggage article names the smaller Unitree R1, while a second heise article the next day, on Unitree's G1 working at the Japanese airport, names the Unitree G1 and says GMO Internet Group is the implementation partner. We could not open JAL's own release, so we cannot say which model, or whether both, are in use.
The second report adds two points that matter more than the model name:
- JAL stressed that safety-relevant tasks will stay with human staff; the robots work alongside ground crews, not in place of their safety duties.
- The tasks were prepared in simulation. Engineers built a digital twin of the work from motion-capture and video data and trained the robot with reinforcement learning before it worked on the apron.
This is a back-of-house trial. The robots work airside with trained staff, not among passengers in the terminal. That is a very different risk profile from a robot greeting the public.
Keeping the robots running
In September 2026, GMO launched a mobile repair service for humanoid robots, described by New Atlas in its report on the robot ambulance. Engineers go to the site to repair a faulty robot; if they cannot fix it there, GMO takes it to its own facility and supplies a replacement. The article links the service to GMO's earlier Haneda work but gives no failure rates or trial results. It is, however, a public sign that the operators expect hardware faults and plan for spare units.
Osaka Metro: a talking humanoid at the ticket gates
Osaka Metro, KDDI and the robot developer AVITA ran a public field trial of a customer-service humanoid from 31 August to 15 September 2026. According to AVITA's announcement of the station guidance trial, the robot stood at Umeda Station (outside the North Gate) until 6 September and then at Tanimachi-kyuchome Station from 9 to 15 September, from 11:00 to 20:00 each day.
The robot answered questions about routes and transfers, suggested nearby attractions and offered a QR code so passengers could take route information onto their phones. AVITA says its dialogue runs on Google's Gemini model and that the system analyses speaking speed, intonation and emotion in real time, and uses eye contact, facial expressions and gestures. It supports several languages; Kyodo's report, republished by NewsOnJapan, put the figure at around 60.
The stated goals are modest and explicit:
- AVITA says the trial was to assess passenger acceptance and operational problems, with mobility and physical tasks to be explored only later. The robot did not move around the concourse.
- Osaka Metro told Kyodo that it wanted to see whether handing information duties to a robot could free station employees for other work, including monitoring passenger safety.
- The operator also pointed to the rising number of international visitors, for whom multilingual help could make assistance easier.
Kyodo quoted passengers who found the robot useful. One said it was easier to ask a robot a casual question than busy station staff. These are individual comments, not survey results; no acceptance figures from the trial have been published. Real-time emotion analysis of passers-by also raises privacy questions that any operator copying the format would need to address.
Hong Kong: a patrol humanoid falls down a staircase
The most instructive incident came from Hong Kong International Airport. On 1 August 2026, a humanoid security patrol robot fell down an emergency-exit staircase in Terminal 1 during a test run, and both its arms came off on impact, VnExpress International reported. The Airport Authority Hong Kong confirmed the incident and said staff of the operations contractor were supervising the robot at the time.
According to the same report:
- The fall happened in a non-public area and nobody was hurt.
- At least two humanoid robots were in the programme, and the robot had been in trials for about four months without problems beforehand.
- The maker and model were not named, and no cause has been published.
The lesson is not that humanoids cannot patrol. It is that stairs, level changes and edges are among the hardest hazards for walking robots, and that months of normal operation do not rule out a sudden failure. Had the staircase been in a public area, a falling robot of this size could have injured someone. Our overview of documented humanoid robot incidents gives further examples.
An older reference point: SPENCER at Schiphol
The current trials are not the first. From 2013 to 2016, the EU-funded SPENCER project, coordinated by the University of Freiburg with KLM as end-user, developed a robot to guide transfer passengers at Amsterdam Schiphol, according to the European Commission's CORDIS project record. The EU contributed about €3.2 million. Its research focus is telling: detecting individuals and groups, modelling social behaviour and navigating politely through busy terminals. A decade later, crowd perception and navigation are still the problems that keep most public-facing transport robots stationary or behind the scenes.
What the three trials have in common
Put side by side, the 2026 trials show a consistent pattern:
- Physical work stays away from passengers. JAL's robots handle baggage and equipment airside; Hong Kong's patrol tests were in a non-public area when the fall happened.
- Public-facing robots stay put. Osaka's humanoid talked to passengers but did not walk through the station, which removes most collision and fall risk in a crowd.
- Humans remain responsible. JAL keeps safety-relevant tasks with staff, Hong Kong's robot was supervised, and Osaka framed the robot as a way to give staff more time for safety work.
- Results are not yet public. No operator has published task completion rates, uptime, passenger survey data or cost figures. Statements so far describe intentions and impressions, not measurements.
What this means for event organisers
Transport hubs are a useful stress test for anyone planning a robot at a busy venue, because they combine dense crowds, noise and strict safety duties.
- Separate roles. A robot that greets and informs can work among visitors if it stays in a fixed spot; a robot that walks or carries things needs a controlled area. Our planning guide to safety at public events sets out how to define that boundary.
- Keep robots away from stairs and edges. The Hong Kong fall happened despite supervision and months of trouble-free tests. Ask suppliers how the robot handles level changes and which areas should be fenced off.
- Budget for people. Every trial kept trained staff next to the robot. Plan for a supervisor or operator for the whole run; see who operates the robot and what staff you need.
- Test the conversation in real noise. Osaka's multilingual, AI-driven dialogue is the kind of feature that works in a demo and struggles in a loud hall; our guide to speech, languages and noisy venues explains what to check.
- Ask for a fallback. GMO's repair-and-replace service shows that even large operators plan for breakdowns. Agree in advance what happens if the robot fails on the day.
Frequently asked questions
Which robot is Japan Airlines testing at Haneda Airport?
Reports disagree. heise online named the Unitree R1 in one article and the Unitree G1 in another published the next day, with GMO Internet Group as the implementation partner. We could not access JAL's own announcement to confirm the model. What is consistent is the scope: ground handling tasks such as loading baggage, cabin cleaning and operating ground equipment, from May 2026 until 2028, with safety-relevant tasks staying with human staff.
Did the Osaka Metro humanoid robot walk around the station?
No. According to AVITA, the trial focused on conversation with passengers and on assessing acceptance and operational problems, with mobility and physical tasks left for later phases. The robot stood at a fixed location near the gates at Umeda and then Tanimachi-kyuchome stations between 31 August and 15 September 2026. This approach avoids most of the collision and fall risks a walking robot would face in a crowd.
Was anyone hurt when the robot fell at Hong Kong airport?
No injuries were reported. VnExpress International reported that the humanoid patrol robot fell down an emergency-exit staircase in a non-public part of Terminal 1 on 1 August 2026 during a supervised test run, losing both arms on impact. The Airport Authority Hong Kong confirmed the incident. The robot's maker and the cause of the fall had not been published at the time of the report.
Have any airport or station robot trials published measured results?
Not for the 2026 trials covered here. Japan Airlines, Osaka Metro and the Airport Authority Hong Kong have described tasks, dates and goals, and journalists have quoted individual passengers, but no operator has released task completion rates, uptime figures, survey data or costs. Organisers should treat claims about robot performance in transport hubs as unproven until such data appears.
How this article was made: written by the humanoid4rent editorial team with AI assistance; every factual statement was checked against the sources listed below on 2026-11-16. Spotted an error? Write to info@humanoid4rent.com.
Nguồn tham khảo
Các thông tin trong bài viết này đã được đối chiếu với những trang dưới đây vào ngày viết bài.
- Humanoid robots to handle baggage at Tokyo airport – heise online
- Unitree's humanoid G1 robot works at Japanese airport – heise online
- Japan's first robot ambulance works the 9-1-1 lines for humanoids – New Atlas
- Humanoid Begins Station Guidance Trial in Osaka, Exploring How Robots Can Work Alongside People – AVITA Inc.
- Osaka Metro Trials Humanoid Robot – NewsOnJapan (Kyodo)
- Patrolling robot breaks apart after falling down staircase at world's best airport – VnExpress International
- Social situation-aware perception and action for cognitive robots (SPENCER) – CORDIS, European Commission
Thêm từ blog
Quý khách muốn có robot tại sự kiện sắp tới?
Chỉ cần mô tả một lần về ngày, địa điểm và nhiệm vụ của robot. Các nhà cung cấp sẽ gửi báo giá để Quý khách so sánh trên cùng một tiêu chí.
Gửi yêu cầuQuý vị làm trong lĩnh vực robot? Hãy viết cho blog của chúng tôi. →