Physix FrontierWhen an Industrial Giant Like Mitsubishi Partners with Humanoid Robot Startups: A Bet on the Future or a Compromise with Reality?
Japan's labor shortage is a well-worn topic. According to data from the Ministry of Health, Labour and Welfare, the job-offer-to-applicant ratio in manufacturing exceeded 1.8 in 2024, meaning each job seeker corresponds to 1.8 positions. Aging combined with declining birth rates means factories aren't lacking equipment—they're lacking people. Mitsubishi's bet on humanoid robots this time seems to address the symptoms directly—using bipedal mechanical bodies to fill vacant assembly line jobs.
But as an entrepreneur who has been grinding in warehousing and logistics, my first reaction wasn't excitement, but calculating deployment costs. Humanoid robots have two fundamental flaws:
First, questionable cost-effectiveness. Based on current BOM cost estimates, a humanoid robot capable of autonomous movement, grasping, and operation won't sell for less than 1 million RMB. A regular worker in a Japanese factory earns approximately 250,000 RMB annually; even including social security and management costs, a humanoid robot needs at least four years to break even. More critically, can it run continuously for four years without breaking down or losing efficiency?
Second, scenario generalization capability is overestimated. My experience running SLAM solutions in warehouses is clear: even 2D laser navigation AGVs need repeated calibration and debugging to operate stably in complex dynamic environments. For humanoid robots to complete cross-station transport, assembly, and inspection in factories, the requirements for motion planning and perception fusion increase exponentially. I don't know the details of Highlanders' technical foundation, but globally, companies that can stably mass-produce bipedal walking robots can be counted on one hand.
Mitsubishi's choice, from a business logic perspective, looks more like a strategic investment in future ecological niches rather than short-term cost reduction. Japan's manufacturing industry has long relied on highly customized production lines; switching to flexible automation requires standardized robot bodies. Mitsubishi has core components like motors, reducers, and CNC systems under its umbrella, plus Highlanders' body design capabilities. Once a closed loop is formed, they can output complete solutions to downstream auto plants and electronics factories. This is the traditional heavy industry enterprise's calculation.
They don't lack money; what they lack is a terminal carrier that can sustain the capital story.
In contrast, many people in our startup circle make the mistake of prematurely fetishizing the humanoid form. I remember chatting with a mobile chassis team in 2021; they insisted on going bipedal, thinking wheeled designs weren't sexy enough. After three rounds of funding, their product was still holding onto railings in the lab, while teams doing four-wheel differential chassis had already shipped 200 units and ran in JD.com warehouses for three years. The biggest illusion of humanoid robots is that they mimic human appearance but cannot mimic human dexterity and fault tolerance.
My own judgment is that in warehousing and logistics, humanoid robots will not become mainstream within the next five years. The reason is simple: the ultimate goal of warehouse scenarios is not to make robots look like humans, but to maximize handling efficiency, optimize paths, and minimize costs. In this dimension, Kiva-like shuttles plus collaborative arms already cover picking for most SKUs, and marginal costs are still dropping. Conversely, scenarios like automotive final assembly lines and nuclear power maintenance, which have high hazard/high posture requirements, offer irreplaceability for humanoid robots—for example, needing to climb high, crawl under cars, and tighten screws of various types.
If the collaboration between Mitsubishi and Highlanders can find a sufficiently narrow entry point, such as welding or painting—roles requiring extremely high dual-arm coordination and posing toxicity risks to humans—and deliver a solution with monthly rent under 20,000 RMB capable of running 24/7, then scalable replication becomes possible. Otherwise, it will likely end up like Mitsubishi's previous robot lab projects, becoming just another footnote in press releases.
Here is specific advice for friends like me on the entrepreneurial path:
If your team is working on humanoid robots, spend at least 70% of R&D budget on hardware reliability and cost reduction, not on AI demos. Investors get excited by demos, but clients care about whether you can run continuously on the production line for three months without failure. Look at Boston Dynamics: their Atlas has been developed for over a decade with no commercialization yet, while Spot (quadruped) has sold thousands of units. Form follows function, and function follows cost.
The next five years of this supply chain will likely see "specialized humanoids landing first, general-purpose humanoids continuing to burn cash." The Mitsubishi-Highlanders collaboration will accelerate the former but won't change the latter. Trend predictions are already clear:
Humanoid robots will first achieve small-batch deployment in specific stations in Japanese auto factories, then enter tooling workshops and precision assembly stages around 2028. But this requires Mitsubishi to tolerate three years of no net profit, and Highlanders' founding team to have enough resolve not to be hijacked by tech hype. Otherwise, this collaboration will ultimately just be a painless "R&D expense" in Mitsubishi's financial reports; and for the entire industry, the truly valuable thing has never been the humanoid shell, but the motion control and perception algorithms that allow robots to adapt to non-standard environments like humans.
Original Link: https://www.ithome.com/0/975/389.htm
