Japanese Humanoid Robots in 2026: Every Major Model Compared

NEXTAGE Fillie ranks first because it is the one Japanese humanoid that factories in Japan can order for real work today. Kawada Robotics began building humanoids for the University of Tokyo in 1999, and Fillie, released in 2022, is the compact model of its NEXTAGE line. It has two arms, two cameras in its head and cameras in its hands, and its motors are kept under 80 watts so it can work safely next to people. It does not use AI to decide what to do. Staff teach it each job with a graphical tool called NxProduction, with no coding skills needed, and it then repeats assembly, inspection and boxing tasks. A research version, NEXTAGE Fillie OPEN, runs ROS. It suits factories in Japan with repetitive bench work and too few hands.

Torobo ranks second because it is a full-size research humanoid that a company or lab can use for real force-sensitive work, and its maker keeps showing new skills on it. Tokyo Robotics fitted torque sensors in every joint of its arms and waist, so Torobo feels how hard it pushes. That lets it hammer a nail, saw wood and wipe a table without crushing anything. The wheeled version has 7-axis arms, a 3-axis waist and an omnidirectional base, and stands 1,660 mm tall. Each arm holds 8 kg in its weakest pose and 20 kg at its peak. Its software is built on ROS, and newer Torobo2 videos show autonomous laundry folding and table clearing, while a two-legged Torobo walks with a human-like gait. Its main limit is that it is a development platform, not a finished worker. It suits labs and companies building their own robot skills.

Ena ranks third because it is being built inside real Japanese care homes, with a beta program set to start in late 2026. Its maker, Tokyo startup Enactic, says it works with more than 70 care facilities in Japan on the design. Ena is meant to take on the physical chores around residents so staff can spend more time with people. The list includes folding laundry, collecting used meal trays, tidying rooms, handling soiled items, wiping tables, restocking supplies, serving tea and snacks, and watching over residents. Ena builds on OpenArm, Enactic's open-source two-arm platform, which the company also sells worldwide to robot researchers. The limit is timing: Ena is not yet at work every day, and facilities join through the beta program first. It suits care operators in Japan who want an early seat and a say in how the robot works.

Cinnamon 1 ranks fourth because it is already on duty in a real nursing home, even though its maker admits it can do little so far. Donut Robotics has trialled the two-legged robot at a Charm Care home since July 2026, starting with patrols. The next step is nurse calls: the plan is that when a resident presses the call button, the robot goes to them, takes the request and helps staff respond. Charm Care, which runs more than 100 private nursing homes, has invested 300 million yen in the company. Cinnamon 1 stands 170 cm tall and can follow hand and finger gestures, a patented feature meant for loud places where voice commands fail. Donut Robotics also plans to use it in factories, building sites and security work. It suits care operators and partners willing to help train a young robot on the job.

Kaleido 8.0 ranks fifth because Kawasaki has spent a decade making it tough enough for real disaster and factory sites, but it is not for sale yet. Kawasaki began building humanoids in 2015 and first showed Kaleido at the iREX trade show in Tokyo in 2017. The Kaleido line runs on its own battery, walks on two legs, and has lifted 50 to 60 kg in bench press demos. From 2023 it could correct where its foot lands when knocked, which cuts the risk of falling. The newer Kaleido 9 adds LiDAR mapping for walking on its own and a head-mounted display for remote control. Kawasaki plans first jobs such as patrols and moving parts around 2030, and is talking with car and chip makers. It suits disaster response teams and heavy industry planning for the next decade.

HRP-4 ranks sixth because it was supplied to outside research groups and is still a full walking humanoid, but it dates from 2010. Kawada built the body and Japan's national research institute AIST wrote the motion control software. It stands 151 cm tall and weighs only 39 kg, with 34 degrees of freedom, meaning 34 separate joints, including 7 in each arm for handling objects. Supply to external research groups began in January 2011. Its software runs on OpenRTM-aist and a real-time Linux system, so labs can share code and use the free OpenHRP3 simulator. Its age is the main limit: newer robots walk faster and use modern AI. It suits universities that want a proven Japanese walking platform with a long research record behind it.

Punyo ranks seventh because its ideas are fresh and useful, but it is a research torso with no legs that is not offered for sale. Toyota Research Institute built Punyo to handle large, heavy items the way people do, by hugging them against its chest and arms rather than gripping with outstretched hands. Each arm is covered in 13 air-filled bubbles, each linked to a pressure sensor, so the robot feels where it touches an object. Its hands are soft latex paws with a camera inside that reads how the surface bends. Under the soft skin sit two rigid arms and a waist motor. Punyo learns from human demonstrations and from training in simulation. The name is a Japanese word for something soft, cute and resilient. It suits researchers studying how future home robots might lift and carry safely.

CUE7 ranks eighth because it is a pure research demonstrator, yet it shows how fast Toyota's robot control is moving. The CUE line began as a side project by Toyota employees in 2017 and set world records for basketball shots, including a Guinness World Record for the longest shot by CUE6 in 2024. CUE7 made its debut on April 12, 2026, at a pro basketball game in Tokyo. Toyota cut its weight from 120 kg to 74 kg and moved it onto two wheels that balance like a self-balancing scooter. A mix of reinforcement learning and model predictive control, two ways of teaching robots to plan their moves, lets it move quickly and still shoot with precision. It is not for sale and has no job beyond the court. It suits anyone following how Japanese firms test vision and balance before using them in other machines.