Mitsubishi Humanoid Robots: 1,000 a Month From Idle Kyoto Line

An engine plant gets a new job

Mitsubishi Motors plans to turn unused space at its Kyoto engine plant into a production line for humanoid robots. The robots come from Highlanders, a startup spun out of the University of Tokyo, and the stated capacity target is 1,000 units a month.14 The two companies signed a memorandum of understanding on July 9, 2026. Under it, they will jointly develop humanoids for Mitsubishi's own factories and look into mass-producing Highlanders' robots in Kyoto.511 Mitsubishi had already made a small investment in Highlanders in May and says it will invest more.23

The headline number has drawn the attention. The more interesting part of the deal is how it is built. Highlanders is developing a robot foundation model, and it needs physical robots in real factories to train it. Mitsubishi has an aging workforce, factory floors that are running below capacity, and decades of experience mass-producing machines to automotive standards. On paper, each side has what the other lacks.

What was actually agreed, and when

The coverage agrees on the basics and differs on how firm the plan is. Some outlets wrote it up as settled: Mitsubishi "will begin" mass production in early 2027.13 Closer readings point out that this is an MOU, not a binding contract. Mitsubishi's own wording about early-2027 production uses terms like "explore" and "under study."512 One independent humanoid tracker classifies the deal as an announcement and feasibility plan, not an active pilot or the start of manufacturing. It notes that no production model, customer pilot or ordering route has been documented.12

The timeline also depends on which milestone is meant. Japanese financial coverage of the launch press conference gives a sequence. Production facilities in the idle Kyoto buildings get prepared in the first half of 2027. The 1,000-per-month capacity is reached in the second half of that year.23 Headlines that put "1,000 a month" and "early 2027" side by side merge two different steps.

The robots will be tested first at Mitsubishi. The company plans to start with several dozen units in its own plants, with the Kyoto engine assembly line as a leading candidate. Many of the tasks there involve little change in posture, which makes them a forgiving first test.710 The first jobs mentioned are moving components and engine assembly, with welding and logistics also discussed.9 Highlanders CEO Hiroya Masuoka has stressed material handling and other tasks that put human workers at risk of injury. Sales to other manufacturers depend on the robots proving reliable inside Mitsubishi first.59

Mitsubishi CEO Takao Kato has been open about why the company is doing this. He called labor shortages an "urgent challenge."7 He also acknowledged that the Kyoto site has spare capacity, including floor space, as domestic vehicle output falls. He said Highlanders' speed was the deciding factor in choosing it as a partner.23

Which robot? The spec sheet is muddled

The reporting is most inconsistent about the robot itself. Several English-language outlets describe the HL Human, a prototype Highlanders first showed in 2025, with a 19-degree-of-freedom body and high-output electric actuators.126 Japanese-sourced coverage of the July press event instead centers on "N," which Highlanders unveiled that day as its fourth-generation humanoid. N has five-fingered hands, several vision systems, a microphone array and a current-generation GPU on board. Its control AI generates the next walking movement about 100 times a second.2311

The physical details don't line up either. One trade report puts the robot at about 1.75 meters tall.3 Another analysis describes a 1.4-meter machine with a 3-kilogram maximum payload.21 Elsewhere, the 3-kilogram figure appears as a grip rating for five-finger hands planned for future versions.1 At least one detailed write-up states plainly that neither company has published full specifications for N, including payload, battery life or degrees of freedom.5 The most likely explanation is that the HL Human figures are being attached to a newer robot. Until Highlanders publishes a datasheet, every specific number should be treated as provisional.

One claimed detail stays consistent. N is built to roughly human proportions, so it can use tools, buttons and pedals designed for people without the factory being rebuilt around it.2321

The foundation-model bet behind the hardware

The development that matters most for robotics AI came on the same day as the MOU. Highlanders announced Kepler v1.0, its own foundation model, which serves as the robot's "brain."23 The design is built around a world model that predicts what will happen next in the physical environment. It splits the work into two layers. A "cerebrum" handles understanding and planning tasks, and a "cerebellum" handles fast reactions such as balance and force control.23

One analysis adds more detail, though no other report confirms it. By that account, Kepler is a multimodal model of about 10 billion parameters. It is trained with simulation, teleoperation and real-world robot experience on an in-house compute cluster called HISUI. Fleet data flows back through a pipeline called "Artery" and returns to the robots as over-the-air updates.21 Those specifics should be treated with caution until Highlanders confirms them. The overall direction, however, is clear from the company's own messaging.

Highlanders describes itself as "the data flywheel for the physical world." Its argument is that language models could learn from the internet, but physical intelligence has no equivalent body of data. That data only comes into existence when someone puts machines into the world and owns what they record.22 The company says it builds its own robot bodies mainly to own that feedback loop: one software stack running across many bodies, with every deployed unit acting as a sensor for Kepler.25

Masuoka has listed four things he considers necessary: huge volumes of real-world data, widely deployed high-performance robots to collect it, mass-production know-how to build those robots well, and real worksites where they can operate.23 Highlanders brings the model and the robot. Mitsubishi brings the other three. Mitsubishi's factories become both a customer and a source of training data, and its quality systems become the way to make robots in large numbers. The deal reads less like a supply agreement and more like infrastructure for collecting data. Masuoka has gone further and framed the long-term ambition as building 900,000 robots a year, roughly matching Japan's annual population decline.23

A different shape from other automaker tie-ups

Most partnerships between carmakers and humanoid developers are buyer-supplier deals, such as BMW with Figure or Toyota with Agility. Hyundai owns Boston Dynamics outright, and Tesla builds Optimus itself. Mitsubishi is doing something different: it is manufacturing a startup's design.5 One survey of Japanese companies calls this a contract model, where a startup's machine is mass-produced in an automotive plant. It notes that the arrangement is common in China but new for a listed Japanese company.11 Another commentator compares it to a "fabless" AI startup working with a "foundry-style" automaker.21

The competition explains the urgency. UBTECH has opened a plant in China with capacity for 10,000 humanoids a year, built together with Siemens.16 Figure has reported producing more than 350 of its Figure 03 robots.12 Highlanders, by contrast, has 46 employees including contractors and has been building robots at a small facility north of Tokyo.22 It is also worth noting that Japan's wider industry still looks strongest in components such as reducers, bearings and actuators. Some analysts argue that is where Japanese companies are most likely to win.11

Policy could also help. On July 28, the U.S. FCC added certain foreign-produced mobile robots to its Covered List, which can block authorization of new models.12 One analyst suggests Mitsubishi's U.S. manufacturing and sales presence could help it work around such restrictions.14 That is speculative for a robot that does not yet ship.

The verdict: a credible structure, unproven numbers

The deal's structure makes sense. Linking a foundation-model company to a guaranteed set of real deployments and to automotive-grade manufacturing tackles the two problems humanoid startups most often fail on: data and the ability to build at scale. The numbers have not yet earned the confidence that headlines have given them. Pricing, customers, maintenance systems and parts sourcing are all undecided.11 Mitsubishi has also not confirmed whether N will be used in its current form. Some reports describe equipment installation and safety reviews already underway, but they cite no named sources and should be treated cautiously.8

The signals to watch are concrete. First, whether the MOU becomes a binding production contract. Second, whether Mitsubishi reports data from its first few dozen in-house robots. Third, whether Highlanders publishes verifiable specifications for N and benchmarks for Kepler.1112 If those arrive, the Kyoto line could become Japan's most serious test of whether a robot foundation model can improve through factory work. If they don't, 1,000 a month will remain a target and nothing more.