Toyota (robotics)
Last reviewed
Sources
41 citations
Review status
Source-backed
Revision
v6 · 5,497 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
41 citations
Review status
Source-backed
Revision
v6 · 5,497 words
Add missing citations, update stale details, or suggest a clearer explanation.
Toyota Motor Corporation is a Japanese multinational automotive manufacturer that, through the Toyota Research Institute (TRI) and its Large Behavior Models program, has become a major force in AI-driven robotics, alongside one of the longest corporate robotics histories in the world (dating to the 1970s). TRI, established in 2016 with a $1 billion, five-year commitment and led by roboticist Dr. Gill Pratt, develops Large Behavior Models (LBMs), single foundation models that teach robots hundreds of manipulation skills from human demonstrations, an approach that in 2025 acquired new tasks with up to 80 percent less training data than learning from scratch [13][29]. Toyota's robotics portfolio spans industrial welding robots, musical humanoid robots, teleoperated avatars, home-assistance platforms, AI-driven basketball robots, and a 2024 research partnership with Boston Dynamics to put LBMs on the electric Atlas robot. Its robotics efforts are split between two organizations: the Partner Robot Division at Toyota's headquarters in Toyota City, Japan, and TRI in the United States.
Toyota's approach to robotics has consistently emphasized augmenting human capabilities rather than replacing people. From its earliest industrial robots to the teleoperated T-HR3 humanoid and TRI's household assistance research, the company pursues what it calls "intelligence amplification," using machines to extend human reach, strength, and precision. As Gill Pratt put it, "Our research in robotics is aimed at amplifying people rather than replacing them" [40]. This philosophy distinguishes Toyota from competitors focused primarily on full autonomy and reflects the company's roots in the Toyota Production System, which has long prioritized the collaboration between people and machines.
Toyota's robotics history began in its automobile manufacturing plants during the 1970s. The company developed industrial robots for welding, painting, and assembly operations to improve quality and reduce costs. One of its most significant early achievements was adapting general-purpose industrial robots for production-line arc welding of heavy steel plates and components. An accelerated research and development program was established, and by mid-1974, Toyota engineers had created robots capable of five types of movement: expansion and contraction, vertical motion, body turn, and wrist turn and twist. These movements could be utilized and controlled in any combination [1].
The welding robots delivered dramatic improvements. Welding time was reduced to roughly one-tenth of manual welding time, and the need for highly skilled welders on certain repetitive tasks was eliminated. This development was reportedly the first of its type in the world for regular production-line arc welding [1].
| Aspect | Details |
|---|---|
| Start of industrial robot development | 1970s |
| Primary applications | Arc welding, spot welding, painting, assembly |
| Key innovation | General-purpose robots adapted for production-line arc welding (by 1974) |
| Welding accuracy (modern) | Within one-tenth of a millimeter |
| Automation level (assembly lines) | Less than 8% of work on global assembly lines as of 2017 |
| Philosophy | Manual processes perfected by employees first, then automated by those same employees |
Notably, Toyota's manufacturing philosophy has kept automation levels on its global assembly lines at relatively modest levels compared to some competitors. For at least the last decade, robots have been responsible for less than 8% of the work on Toyota's assembly lines. The company follows a principle where manual processes are first perfected by employees, and only then automated by those same employees, reflecting the broader Toyota Production System emphasis on human judgment and continuous improvement [2].
In February 2026, Toyota signaled a new direction for factory automation when Toyota Motor Manufacturing Canada (TMMC) announced a commercial agreement to deploy general-purpose humanoid robots on its production support operations. See the 2025-2026 developments section below for details [29][30].
In the early 2000s, Toyota expanded its robotics ambitions beyond the factory floor. The company established the Partner Robot Division to develop humanoid robots capable of coexisting with and assisting people in daily life. Toyota's goal was to fuse its industrial robot technology, which it had refined since the 1970s, with automotive, electronics, and intelligence technologies to create robots that could help people in manufacturing, personal mobility, nursing and healthcare, and housekeeping [3].
The Partner Robot program gained worldwide attention at Expo 2005 in Aichi, Japan, where Toyota showcased a series of humanoid robots performing as a musical ensemble. The robots played trumpets, tubas, trombones, French horns, and percussion instruments, demonstrating Toyota's capabilities in fine motor control, artificial lip articulation, and bipedal locomotion.
The most prominent robot in the ensemble was the Version 1 Walking Type, a bipedal trumpet-playing robot nicknamed "Harry." Three units were built, each standing approximately 145 cm tall, weighing around 40 kg, and featuring 31 degrees of freedom. Harry was designed to play the trumpet authentically, a task that requires dexterous fingers, flexible lips, and controlled airflow. The robot used artificial lips and mechanical lungs to produce actual sound from a real trumpet, rather than playing pre-recorded audio [4].
The trumpet-playing robots were designed to express the Japanese concept of "wa" (harmony), presenting robots as friendly companions rather than mere industrial tools. The performance at Expo 2005 was a landmark moment in entertainment robotics and signaled Toyota's serious commitment to humanoid platforms.
The Partner Robot series consisted of five distinct platforms, each using a different mobility system:
| Version | Mobility System | Key Feature |
|---|---|---|
| Version 1 (Walking Type) | Bipedal legs | Trumpet-playing humanoid; 31 DOF |
| Version 2 | Segway-like wheels | Wheeled musical performer |
| Version 3 | Segway-like wheels | Wheeled musical performer |
| Version 4 | Wire suspension system | Suspended performance robot |
| i-Foot | Two-legged mountable pod | Rideable bipedal transport |
The variety of locomotion systems reflected Toyota's experimentation with different mobility approaches for different use cases. Later developments included a violin-playing robot unveiled in December 2007, which featured 17 joints across both hands and arms. The five-foot-tall robot coordinated its mechanical fingers to press violin strings while bowing with its other arm [5].
In July 2009, Toyota demonstrated a running version of the Partner Robot that achieved speeds of up to 7 km/h, taking a step every 340 milliseconds with 100 milliseconds of flight time (no ground contact). However, walking and running were limited to flat surfaces [6].
The T-HR (Toyota Humanoid Robot) series represents Toyota's most advanced humanoid robot program. The "3" in T-HR3 indicates it is the third generation, following the T-HR1 and T-HR2 internal research prototypes. While Toyota has not published extensive details about the first two generations, they served as research platforms that informed the development of key technologies later integrated into the T-HR3.
Unveiled on November 21, 2017, the T-HR3 is a teleoperated humanoid platform designed for safely managing physical interactions between robots and their environments. It is controlled through a full-body teleoperation interface called the Master Maneuvering System (MMS), which maps the operator's movements to the robot in real time while relaying visual and force feedback.
| Specification | Value |
|---|---|
| Height | 1,540 mm (154 cm) |
| Weight | 75 kg |
| Degrees of freedom | 32 (plus 10 fingers per hand) |
| Control method | Master Maneuvering System (teleoperation) |
| Key technology | Torque Servo Modules (Cr-N alloy thin-film torque sensors) |
| MMS weight | 170 kg |
| MMS control axes | 16 |
The T-HR3's fundamental building block is the Torque Servo Module, a compact unit combining a motor, a reduction gear, and a supersensitive torque sensor at every joint. These modules were developed in collaboration with Tamagawa Seiki Co., Ltd. and NIDEC COPAL ELECTRONICS CORP. The architecture enables three core capabilities: flexible joint control (regulating contact forces), whole-body coordination and balance control (maintaining stability during collisions), and real remote maneuvering (providing intuitive operator control with haptic feedback) [7].
In November 2018, Toyota partnered with NTT DOCOMO to demonstrate remote control of the T-HR3 over a 5G mobile network across a distance of approximately 10 kilometers, confirming that the robot could perform precise force-transmission tasks at levels comparable to wired connections [8]. An upgraded version shown at IREX 2019 and CES 2020 featured lighter arm and leg components, a new hand controller for finer finger manipulation, and smoother bipedal walking.
For a full account of the T-HR3's design, Master Maneuvering System, and development timeline, see the dedicated T-HR3 article.
As a top-level Olympic sponsor, Toyota deployed multiple robot systems at the Tokyo 2020 Olympic and Paralympic Games (held in 2021 due to the COVID-19 pandemic). The most visible were the Miraitowa and Someity mascot robots, robotized versions of the official Tokyo 2020 mascot characters.
The mascot robots applied teleoperation and force-feedback technology directly derived from the T-HR3 program. Software was developed to coordinate movements in real time and to compensate for the physical differences between human operators and the mascots' proportions (extremely large heads and short limbs compared to the T-HR3). Each mascot had sensors embedded in the soles of its feet that calculated the center of gravity every 1/1,000 seconds, sending signals to joint motors to produce lifelike movements. Cameras mounted on their foreheads allowed the robots to recognize people nearby and react to them [9].
Miraitowa and Someity welcomed athletes and guests at Games venues, performing gestures such as shaking hands, waving, and displaying a variety of facial expressions. Remote users could experience force feedback while interacting through the mascot avatars, allowing them to feel as if they were physically present at the venue [10].
Toyota also deployed the Human Support Robot (HSR) for spectator assistance and delivery support robots at the Games, demonstrating the breadth of the company's robotics portfolio.
The Human Support Robot is a compact mobile manipulator developed by Toyota's Partner Robot Division to assist elderly and disabled individuals. First introduced in 2012, the HSR features a cylindrical wheeled base with a telescoping body and a folding arm equipped with a two-fingered gripper.
| Specification | Value |
|---|---|
| Height (adjustable) | 82 to 131 cm (2.7 to 4.3 ft) |
| Weight | 32 kg (70 lb) |
| Arm length | 76 cm (2.5 ft) |
| Grip capacity | Up to 1.2 kg (2.6 lb) |
| Degrees of freedom | 8 (body) plus pan-tilt head |
| Maximum speed | 3 km/h |
| Obstacle clearance | Up to 0.8 cm bumps; slopes up to 5 degrees |
| Software platform | ROS (Robot Operating System) |
| Control methods | Voice command, tablet PC, remote operation |
The HSR can pick objects from the floor, retrieve items from shelves, suction up thin objects, and open curtains. It can also be operated remotely, relaying the operator's face and voice in real time. Since 2015, Toyota has offered the HSR to researchers worldwide as an open innovation platform. As of March 2025, 67 research institutions from 14 countries participate as members of the HSR research community, collaborating on topics including Robotics Foundation Models [11][12].
The HSR and T-HR3 represent complementary approaches within Toyota's strategy: the HSR focuses on semi-autonomous domestic assistance, while the T-HR3 explores full-body teleoperation and avatar-style remote presence.
The Toyota Research Institute was established in January 2016 with a $1 billion investment over five years. Headquartered in Los Altos, California, with additional offices in Cambridge, Massachusetts, TRI focuses on artificial intelligence, automated driving, materials science, and robotics. The institute was founded to accelerate Toyota's capabilities in areas where software and AI intersect with physical systems [13].
TRI is led by CEO Dr. Gill Pratt, a roboticist who previously managed the DARPA Robotics Challenge and Neuromorphic Computing programs from 2010 to 2015. Under Pratt's leadership, TRI pursues an "intelligence amplification" approach, where machine learning and AI technologies augment human capabilities rather than replace them. Other key leaders include Dr. Russ Tedrake, Senior Vice President of Large Behavior Models, who also holds the Toyota Professorship at MIT in the departments of Electrical Engineering and Computer Science, Mechanical Engineering, and Aero/Astro [14][15].
TRI's robotics division addresses the challenges of aging populations in Japan and other countries, focusing on household assistance. Key research areas include:
| Research Area | Description |
|---|---|
| Behavior learning from demonstration | Teaching robots to generalize task knowledge from human demonstrations |
| Continuous learning | Enabling robots to learn and share skills from real-world experience |
| Whole-body tactile sensing | Developing comprehensive sensing for physical interaction with environments |
| Simulation-based testing | Using simulation to test systems across diverse scenarios |
| Soft robotics | Compliant manipulation using Soft Bubble Grippers and Punyo platform |
| Ceiling-mounted robots | Gantry robot concept for domestic tasks |
TRI has developed the Soft Bubble Gripper, a manipulator that uses air-filled, elastic latex bubbles for compliant gripping. Inside each bubble is a low-cost Time-of-Flight depth sensor and IR camera that uses vision to "feel" what the gripper is holding. The bubbles detect shapes, forces, and slip conditions, allowing the robot to perform tasks like precisely placing mugs in a dishwasher or sorting recyclables by touch rather than vision. In May 2021, TRI open-sourced the design files at punyo.tech so any research institution could build its own gripper [16].
Building on this work, TRI unveiled Punyo in 2024, a torso-up humanoid research platform whose arms and chest are covered in compliant materials and tactile sensors. The name comes from the Japanese word "punyo" (meaning something soft, cute, and resilient). Punyo's arms feature 13 individually pressurizable air-filled bladders per arm, and the platform is designed for whole-body manipulation of bulky objects, using its arms and chest together rather than relying solely on hand grasps [17].
TRI has also explored a ceiling-mounted "gantry robot" concept, where a robotic system descends from an overhead framework to perform household tasks such as loading dishwashers, wiping surfaces, and clearing clutter. By traveling on the ceiling, the robot avoids the challenges of navigating floor clutter and cramped spaces. When not in use, it tucks itself up out of the way. While the concept faces practical challenges around home infrastructure, it represents TRI's creative approach to domestic robotics [18].
One of TRI's most influential contributions to the robotics field has been the development of Diffusion Policy and the concept of Large Behavior Models (LBMs).
On September 19, 2023, TRI announced a breakthrough generative AI approach to teaching robots new behaviors. Developed in collaboration with Professor Shuran Song's group at Columbia University, Diffusion Policy represents a robot's visuomotor policy as a conditional denoising diffusion process. The approach generates robot actions conditioned on sensor observations of human movement and natural language, enabling rapid behavior learning from demonstration [19]. Announcing the work, Gill Pratt said, "This new teaching technique is both very efficient and produces very high performing behaviors, enabling robots to much more effectively amplify people in many ways" [40].
The key innovation is that Diffusion Policy is naturally stable to train, can learn from demonstrations that achieve goals in different ways, and is well suited to high-dimensional continuous action spaces. Using this approach, TRI taught robots more than 60 dexterous skills, including pouring liquids, using tools, and manipulating deformable objects, without writing a single line of new code. The only change required was supplying the robot with new training data. The Diffusion Policy paper was published at Robotics: Science and Systems (RSS) 2023 and became a community favorite for the Best Paper Award [19][20].
TRI frames Diffusion Policy as a step toward building Large Behavior Models for robots, analogous to Large Language Models (LLMs) that have transformed conversational AI. The LBM vision is a single foundation model that can learn hundreds of manipulation tasks and use prior knowledge to acquire new skills with dramatically less training data.
TRI's custom-built robot platform for LBM research features dual-arm manipulation with a special focus on haptic feedback and tactile sensing. The institute set ambitious targets: hundreds of new skills by the end of 2023, reaching 1,000 by the end of 2024. In 2025, TRI published results showing that a single LBM can learn hundreds of tasks and acquire new skills with 80% less training data than training from scratch [21].
On July 11, 2025, TRI published a detailed study, "A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation," quantifying these gains. The work reported that a single pretrained LBM could acquire new tasks with up to 80 percent less data (a 3 to 5 times reduction in challenging settings that require robustness to environmental variation) compared with training a policy from scratch. The underlying models used a diffusion transformer architecture that processes visual, proprioceptive, and text inputs, and they were trained on nearly 1,700 hours of robot interaction data drawn from both simulation and the real world. TRI evaluated performance across 29 tasks using more than 47,000 simulation rollouts and roughly 1,800 real-world trials, and found that performance improved steadily as the amount of pretraining data increased [29][30].
On October 16, 2024, Boston Dynamics and TRI announced a research partnership to accelerate the development of general-purpose humanoid robots. The partnership combines TRI's Large Behavior Models with Boston Dynamics' electric Atlas robot platform. Scott Kuindersma, Senior Director of Robotics Research at Boston Dynamics, and Russ Tedrake co-lead the Boston-based collaboration [22]. Explaining the rationale, Tedrake said, "One of the main value propositions of humanoids is that they can achieve a huge variety of tasks directly in existing environments, but the previous approaches to programming these tasks simply could not scale to meet this challenge. Large Behavior Models address this opportunity in a fundamentally new way: skills are added quickly via demonstrations from humans, and as the LBMs get stronger, they require less and less demonstrations to achieve more and more robust behaviors" [41].
In August 2025, the partnership demonstrated a Large Behavior Model powering the Atlas humanoid robot. In a video released by both organizations, Atlas performed a long, continuous sequence of complex tasks, using whole-body movements such as walking, crouching, and lifting to accomplish packing, sorting, and organizing tasks. This represented a key step toward general-purpose humanoid capabilities [23]. In the demonstration, a single language-conditioned LBM had direct control of the entire robot and treated the hands and feet almost identically, rather than coordinating separate locomotion and manipulation controllers. The same policy carried out tasks ranging from simple pick-and-place to more involved actions such as tying a rope, flipping a barstool, unfurling and spreading a tablecloth, and manipulating a roughly 22 pound (10 kg) car tire, with new capabilities added through training data rather than hand-written code [31][32].
The CUE series is a line of AI-powered basketball-playing humanoid robots developed by Toyota employees as a volunteer project during their free time. Originally conceived as an engineering challenge, CUE has evolved through several generations and holds two Guinness World Records.
| Generation | Debut Date | Venue | Key Achievement |
|---|---|---|---|
| CUE | March 28, 2018 | Alvark Tokyo home games | First appearance; demonstrated basketball shooting |
| CUE2 | November 24, 2018 | Alvark Tokyo home games | Improved shooting accuracy |
| CUE3 | April 10, 2019 | Alvark Tokyo home games | Guinness World Record: 2,020 consecutive free throws (May 17, 2019) |
| CUE4 | November 16, 2019 | Alvark Tokyo home games | Competed in B.League All-Star three-point shootout (January 18, 2020); won BREAK THE BORDER AWARD |
| CUE5 | July 25, 2021 | Tokyo Olympic Games | Demonstrated half-court shots and dribbling during halftime |
| CUE6 | 2024 | Alvark Tokyo home games | Guinness World Record: farthest shot by a humanoid robot, 24.55 m / 80 ft 6 in (September 26, 2024) |
| CUE7 | April 12, 2026 | Toyota Arena Tokyo (Alvark Tokyo home game) | Full model redesign; dribbling and free movement around the court |
CUE is not simply a mechanical catapult. The robot uses artificial intelligence to recognize patterns and correct its posture, arm position, and shot strength for variables in real time, similar to how a human player adapts. CUE3's free-throw record of 2,020 consecutive baskets took six hours and 35 minutes to complete. Videos of CUE3's performance accumulated over 30 million views worldwide [24][25].
CUE5 appeared during the Tokyo 2020 Olympic Games basketball halftime shows, demonstrating half-court shots and dribbling skills to a global audience. CUE6 then secured the project's second Guinness World Record on September 26, 2024, sinking a shot from 24.55 meters (80 feet 6 inches) at a facility in Nagakute, Aichi, Japan [25].
CUE7, the seventh-generation robot, debuted on April 12, 2026 at a B.League Alvark Tokyo home game held at the newly opened Toyota Arena Tokyo. Toyota described it as a full model change rather than an incremental update. Where earlier CUE robots stayed in a fixed position to shoot, CUE7 can move freely around the court and dribble the ball with smoother, more human-like motion before shooting [33]. According to coverage of the unveiling, the robot's body weight was reduced from about 120 kg to 74 kg and it adopted an inverted two-wheel base for mobility, while its shooting and movement were driven by a hybrid control system that combines reinforcement learning with model predictive control rather than fully pre-programmed motions [34][35].
In 2018, Toyota established Toyota Research Institute - Advanced Development (TRI-AD) in Tokyo as a joint venture with Denso and Aisin, representing a $2.8 billion investment to develop self-driving software. In January 2021, TRI-AD was reorganized and expanded into Woven Planet Holdings (later renamed Woven by Toyota, Inc.), which oversees autonomous driving, connected mobility, and robotics technology [26].
Woven by Toyota also manages Woven City, a purpose-built smart city under construction at the base of Mount Fuji in Susono, Shizuoka Prefecture. Woven City is designed as a living laboratory for testing autonomous vehicles, robotics, smart home technologies, and AI in a real urban environment. The project reflects Toyota's broader ambition to integrate robotics into the fabric of daily life [26].
In April 2021, Woven Planet acquired Lyft's Level 5 autonomous driving division for $550 million, and later that year acquired Renovo Motors, a developer of automotive operating systems. Woven Capital, the group's investment arm, manages an $800 million global fund [27].
Woven by Toyota develops Arene, a vehicle software platform intended to turn Toyota cars into software-defined vehicles. The platform is built around three core packages: the Arene Software Development Kit (SDK), which lets developers build modular, reusable on-vehicle software components; Arene Tools, enterprise tools for visualizing, testing, and validating software in virtual environments to reduce reliance on physical prototypes; and Arene Data, a consent-based data infrastructure that supports over-the-air updates and continuous improvement for autonomous driving, advanced driver assistance, and personalized in-car experiences [36]. In May 2025, Woven by Toyota announced that Arene would make its first customer-facing appearance in Toyota's all-new RAV4, becoming the first Toyota vehicle delivered with features developed on and powered by Arene. In that vehicle, the Arene SDK was used to build the cockpit voice agent and center display of the new multimedia system, along with the latest iteration of Toyota Safety Sense [36].
Toyota's robotics strategy spans four core areas that the company has outlined for its research and development through 2025 and beyond:
| Focus Area | Primary Organization | Approach |
|---|---|---|
| Household assistance | TRI / Partner Robot Division | HSR platform, LBMs, Punyo, gantry robots |
| Nursing and healthcare | Partner Robot Division | HSR, teleoperation, assistive devices |
| Manufacturing | Toyota manufacturing plants | Industrial welding, painting, and assembly robots |
| Personal mobility | Partner Robot Division | i-Foot, mobility concepts |
Across all these areas, Toyota emphasizes "Mobility for All," a vision of using robots and AI to extend mobility and independence to people regardless of age or physical ability. The company frames its T-HR3 teleoperation work as "virtual movement," where an operator's body is projected into a remote space through a physical avatar, unlike virtual reality which simulates environments digitally [28].
Toyota's sustained investment in robotics, from its 1970s welding robots through the Partner Robot program, the $1 billion TRI initiative, and the 2024 Boston Dynamics partnership, reflects one of the longest and most comprehensive corporate commitments to robotics in the world. The company's willingness to invest across basic research, entertainment robotics, practical home-assistance platforms, and frontier AI positions it as a major force in shaping the future of human-robot interaction.
Through 2025 and the first half of 2026, Toyota's AI and robotics work advanced on several fronts: TRI quantified the data efficiency of its Large Behavior Models, the Boston Dynamics collaboration produced a whole-body LBM demonstration on Atlas, Toyota contracted commercial humanoid robots for a North American plant, Woven City opened as a live test course and began deploying in-house AI models, and the CUE basketball project reached its seventh generation. The LBM, Atlas, and CUE7 milestones are described in their respective sections above; the following covers the manufacturing deployment and the Woven City developments.
On February 19, 2026, Agility Robotics announced a commercial agreement with Toyota Motor Manufacturing Canada (TMMC) to deploy seven Digit humanoid robots at TMMC's plant in Woodstock, Ontario, which builds the RAV4. The agreement followed a pilot that ran for roughly one year and began with three units. It was structured as a robots-as-a-service arrangement, in which Toyota pays a subscription that bundles the hardware, maintenance, and software updates rather than buying the robots outright. The Digit robots are assigned to production support work, unloading totes of auto parts from an automated warehouse tugger, a repetitive material-handling task. Reporting described the deal as among the first commercial deployments of humanoid robots in the Canadian automotive industry. TMMC is Toyota's largest manufacturing operation outside Japan, with assembly plants in Cambridge and Woodstock, Ontario [29][30]. TMMC president Tim Hollander said the company chose Digit "to improve the team member experience and further increase operational efficiency," and Agility Robotics chief technology officer Pras Velagapudi noted that the cost of deploying a humanoid can exceed the price of the robot itself, with AI tools used to bring that deployment cost down [30].
Toyota Woven City officially launched on September 25, 2025, in Susono City, Shizuoka Prefecture, as a real-world test course for mobility. Several Toyota Group employees and their families began living there as the first residents, called "Weavers," with Phase 1 ultimately expected to house about 300 residents. Toyota said it planned to begin welcoming general visitors in fiscal year 2026 [37]. Woven by Toyota frames the city's activity around a concept it calls "Kakezan" (Japanese for multiplication), meaning new value created by combining Toyota's mass-production experience, Woven by Toyota's software work, and the contributions of outside "Inventors" and partners [37].
On April 22, 2026, Toyota and Woven by Toyota unveiled a set of in-house AI technologies being deployed across Woven City. The central system is the Woven City AI Vision Engine, a large-scale vision-language foundation model that combines visual, behavioral, and environmental data from sources such as camera feeds, mobility systems, and user inputs to identify patterns, detect potential risks, and coordinate action across connected systems. The company said the model ranks among the world's leading vision-language models, citing the MVBench leaderboard, and that it was being run in a proof of concept with UCC Japan Co., Ltd. Built on top of it, the Woven City Integrated ANZEN System ("anzen" means safety in Japanese) combines the AI Vision Engine with two further components: Woven City Behavior AI, which interprets and predicts human behavioral patterns, and Woven City Drive Sync Assist, which provides driving assistance based on driver needs and surrounding conditions. By analyzing camera data from vehicles and traffic signals, the system aims to anticipate movement and relay that information to pedestrians and drivers. Supporting data infrastructure was introduced under the names Woven City Infra Hub and Woven City Data Fabric. Toyota also said that chairman Akio Toyoda contributed to an internal tool called "Akio Toyoda AI," intended to reflect his approach to leadership and decision-making and to encourage AI adoption inside the organization [38][39].