A fire hose, a fallen shelf and a rescue drill
Kawasaki Heavy Industries' latest Kaleido humanoid handled a fire hose, lifted a toppled shelf weighing 30 kilograms and carried a toy cat to safety at the 2025 International Robot Exhibition in Tokyo. The staged rescue brought three demanding tasks together: following commands, moving a heavy obstacle and finding a stand in for a victim.
Contents
- A fire hose, a fallen shelf and a rescue drill
- Why Kawasaki is building around the human body
- Who controls the robot when work becomes difficult?
- An office robot that can recognize clutter
- Fanuc turns a spoken request into a dice arrangement
- Movement, dancing and mobility draw an audience
- How AI connects seeing with action
- Touch helps robots handle fragile objects
- Labor needs meet cost and reliability limits
- Inside the four day Tokyo showcase
- Key Points
The demonstration was among the attractions drawing crowds at Tokyo Big Sight in Koto Ward, where a record 673 companies and organizations gathered for the biennial exhibition. The 26th edition opened Wednesday with a schedule running through Saturday. Organizers expected attendance to exceed 150,000 over the four days.
Under the theme Sustainable Societies Through Robotics, exhibitors presented machines intended to work with people in disaster zones, factories, offices and public spaces. Humanoids were a headline attraction, alongside industrial arms, remotely operated equipment, dexterous hands and personal mobility devices.
Kaleido's rescue sequence captured the practical ambition behind the displays. A machine that can handle a hose or move a shelf could take on physically demanding work while a person directs the operation from a safer position. Kawasaki also demonstrated less dramatic routines, including raking leaves and taking out trash, connecting emergency work with familiar household chores.
Across the exhibition, the same question took different forms: how can robots use their bodies, cameras and software to perform useful work in places already built for people? Some demonstrations focused on strength, others on delicate handling, spoken commands or keeping balance while moving. Many combined automatic actions with guidance from a human operator.
Why Kawasaki is building around the human body
Kawasaki engineers continue to favor a robot with human proportions because it can wear standard protective equipment and handle hardware already used by firefighters and rescue crews. Two arms, two legs and a torso give a machine access to spaces and tools designed around the human body.
Doors, stairs, valves and hand tools illustrate the advantage. A robot of suitable size can pass through a doorway, climb steps, twist a valve or shoulder a hose without requiring each building or piece of equipment to be redesigned. Using existing infrastructure can reduce the cost of deployment.
Kawasaki Managing Executive Officer Matsuda Yoshimoto described the company's aim as providing support in dangerous places while allowing robots and people to work together.
Robots can perform tasks at disaster hit areas where people cannot step in. Humanoids can also use the same tools as humans. I want to build robots that can work alongside humans.
Kaleido's staged blaze and rescue drill put that approach into a visible sequence. The robot tracked commands, extinguished the mock fire with a hose, lifted the shelf and located the toy cat before carrying it to safety. Each action involved equipment or obstacles that a human rescue team might need to handle.
Who controls the robot when work becomes difficult?
Most humanoids on display combined autonomy with teleoperation. Autonomy means the machine performs some actions through its own software. Teleoperation means a person controls it from a distance. An operator can guide a task when precision or judgment is needed, while onboard systems handle other parts of the movement.
Cameras and software help the robot recognize objects, plan where to put its feet and maintain balance in real time. The person can remain responsible for decisions while the machine carries weight, reaches into danger or repeats physically demanding steps without fatigue.
Man Machine Synergy Effectors demonstrated this approach by combining a humanoid upper body with an aerial work platform. A worker wearing virtual reality goggles could see from the robot's viewpoint and control its arms to tighten fittings or clear debris.
The setup is aimed at work such as railway maintenance and timber felling. Height, heavy components and unpredictable terrain can expose crews to danger. Remote control allows an experienced technician to apply skill and discretion while the robot takes on the physical exposure.
The exhibition's use of mixed control also offered a practical route toward broader deployment. Automatic systems can handle routine steps, with human operators taking over when a task calls for a decision that depends on the situation.
An office robot that can recognize clutter
Yaskawa Electric and SoftBank presented a setting showing what an office of the near future might look like. Their physical AI robot scanned the room through a camera, recognized clutter within seconds and moved to clean and return items to their places.
Physical AI describes software that connects an understanding of the surroundings with actions in the real world. In the office demonstration, the robot needed to do more than identify an object in an image. It also had to turn that recognition into movement and handling.
The sequence showed how a robot could respond to simple commands and tidy a shared space after a meeting or at the end of a shift. Such work requires the machine to deal with the arrangement it encounters, rather than repeat an action at exactly the same spot every time.
Kawasaki's demonstrations of raking leaves and taking out trash extended the focus to domestic tasks. Together, the office and household routines presented a possible progression from specialized missions to everyday assistance.
Other displays included customer service uses, such as taking orders or carrying items through a crowd. Preparing treats at a booth provided another demonstration of precise manipulation, where the usefulness of a machine depends on controlled handling as much as strength.
Fanuc turns a spoken request into a dice arrangement
Fanuc showed an industrial arm equipped with AI that could understand spoken instructions and use a camera to locate particular objects. In one demonstration, a visitor asked it to pick up dice and arrange them so that each displayed the number five.
The system detected three dice, grasped them one at a time, rotated them to show five and placed them in position. The task combined listening, seeing, gripping and changing an object's orientation.
Voice and vision can reduce the amount of programming needed for such work. Instead of specifying every movement, a person can state the desired result, while the system identifies the objects and works out the actions needed to arrange them.
For people without specialist programming skills, that can make complex tasks easier to request. The dice demonstration also showed how useful robotic work can involve several small decisions in sequence: which object to pick up, how to hold it and which way to turn it before setting it down.
Movement, dancing and mobility draw an audience
GMO Internet Group presented a humanoid dancer that jumped, turned and kept time with music on a small stage. The company also worked with a Unitree robot to stage a dance off, drawing attention to agility and balance.
Dance routines brought movement quality into view. A humanoid must control its body throughout a jump or turn, just as it must maintain balance while reaching, lifting or using both hands during a practical task.
Tomohiro Uchida of GMO linked recent improvements in motion to better control algorithms and actuators, the mechanisms that produce movement. He also connected the development of humanoids with a declining labor force.
Robots have become much more fluid in their movements in recent months. Humanoid robots can help fill gaps in the declining labor force.
Kawasaki displayed its personal mobility device Corleo, while Unitree exhibited its Go2 W robot dog. The Go2 W adds wheels to legs for speed and stability, presenting a different way of combining movement systems.
The entertainment and mobility displays sat alongside demonstrations with more direct workplace uses. Across them, exhibitors showed robots crossing obstacles, keeping balance and moving through spaces where people may also be present.
How AI connects seeing with action
Machine learning helps robots recognize a scene, separate individual objects within it and anticipate how those objects will move when pushed or lifted. Separating objects matters because a camera image contains many things at once, while a robot often needs to handle just one.
Pretrained vision models support perception. These are models that have already learned patterns used to interpret images. Policy models then connect images and goals with actions by the robot's hands or feet.
Low level control remains essential. A humanoid using both hands must still keep its body balanced, so recognizing an object and deciding to lift it are only parts of the operation. The machine also needs to coordinate the movements that hold it upright.
Developers at the exhibition described these connected abilities as physical AI: software that understands and acts in the physical world. The office cleaning robot and Fanuc's dice arrangement illustrated different forms of that connection, from recognizing clutter to turning a spoken instruction into a series of movements.
The combination of perception, movement and human supervision points toward robots handling a broader range of work, with people setting goals and managing exceptions. The exhibition's repeated use of remote control showed how much of that work still involves cooperation between a machine and its operator.
Touch helps robots handle fragile objects
PaXini displayed a new generation of tactile sensors and a dexterous hand, part of an effort to give robots feedback closer to human skin. Cameras help identify where an object is; touch helps a machine understand what happens when its fingers make contact.
Tactile sensors measure very small forces and shear, the force acting along a surface. That information can reveal whether an object is slipping in the grip or whether a surface is fragile, allowing handling to depend on more than appearance alone.
Vendors at several booths demonstrated hands picking up a test tube, cradling a sphere and stacking small cubes without cracking or dropping them. These tasks demand careful control because gripping too weakly can let an object fall, while excessive force can damage it.
The tactile displays added another layer to the exhibition's focus on useful work. A rescue robot may need strength to move an obstacle, while a robot handling a small object needs sensitivity. Both depend on matching the action to the task.
Labor needs meet cost and reliability limits
The 673 participating companies and organizations came from Japan, China, the United States and other regions. Their presence reflected intense investment and competition in humanoids and flexible industrial systems, including movement, object handling and the AI linking perception with action.
Exhibitors stressed collaboration with people, with machines taking on repetitive, dangerous or physically heavy work. Demand is rising as many economies face shrinking workforces and cities prepare for earthquakes, floods and fires.
Cost, endurance and reliability remain central to wider use. Batteries need to power arms and legs through a full shift, and hardware must withstand dust, water and changes in temperature. A machine intended for a disaster zone faces different conditions from a carefully arranged exhibition stage.
Safety certification and training programs will also take time before robots can work in warehouses, stations and clinics without creating new hazards. The exhibition gave no timetable for broad deployment. Many exhibitors presented a combination of autonomy for routine actions and teleoperation for judgment calls as a bridge toward more extensive use.
Kaleido's rescue tasks, the remotely controlled aerial platform and the office cleaning sequence showed different ways to divide responsibilities between people and machines. Their shared aim was practical cooperation: people provide direction and experience, while robots contribute physical effort, reach or exposure to danger.
Inside the four day Tokyo showcase
Organizers divided the International Robot Exhibition into an industrial robot zone and a service robot zone. Welding and assembly cells featured on the industrial side, while service displays covered personal assistance, cleaning, agriculture, space and public safety.
The event's schedule and visitor arrangements were:
- Wednesday: the 26th International Robot Exhibition opened at Tokyo Big Sight in Tokyo's Koto Ward.
- Wednesday through Saturday: the four day program brought together a record 673 companies and organizations.
- Saturday: the exhibition's scheduled final day.
Admission was priced at 1,000 yen, with free entry for junior high school students and younger children. Organizers expected more than 150,000 visitors across the four days, a forecast rather than a final attendance count.
The show was designed for industry professionals and curious families alike. Many booths invited visitors to interact with displays and offered demonstrations suitable for photographs. The mix ranged from welding and assembly to rescue drills, delicate handling and dance performances, placing familiar entertainment beside work that could support people in demanding settings.
Key Points
- A record 673 companies and organizations participated in iREX 2025 at Tokyo Big Sight, under the theme Sustainable Societies Through Robotics.
- Kawasaki's latest Kaleido used a hose on a staged fire, lifted a shelf weighing 30 kilograms and carried a toy cat to safety.
- Yaskawa Electric and SoftBank demonstrated an office robot that recognized clutter within seconds and moved to tidy it.
- Man Machine Synergy Effectors showed remote control through virtual reality goggles for tasks aimed at railway maintenance and timber felling.
- Fanuc's AI arm followed a spoken request to arrange three dice with the number five showing.
- GMO presented humanoid dancing and a dance off with a Unitree robot. Kawasaki displayed Corleo, and Unitree showed its Go2 W robot dog.
- PaXini presented tactile sensors and a dexterous hand, while other displays showed delicate handling of test tubes, spheres and cubes.
- Cost, battery endurance, durability, safety certification and training remain central to wider deployment.
- The exhibition opened Wednesday and was scheduled through Saturday, with attendance expected to exceed 150,000.
- Admission cost 1,000 yen, with free entry for junior high school students and younger children.






