1. [para. 1][para. 2] The robotics industry is discovering that humanoid robots are far from ready for unpredictable domestic environments, so the controlled, standardized settings of industrial manufacturing—particularly auto factories—have become the primary testing ground for ambitious humanoid-robot makers.
2. [para. 3][para. 4] Auto plants were initially hailed as ideal laboratories. While stamping, welding, and painting are almost fully automated by traditional machinery, final assembly still depends heavily on human workers. Struggling with volatile demand and high labor costs, automakers hoped humanoids could serve as a flexible workforce. Pilot programs emerged globally: BMW partnered with Figure Robotics in the U.S., while Geely, Xiaomi, Changan Auto, and GAC in China announced humanoid deployments on their factory floors.
3. [para. 5][para. 6] However, the enthusiasm proved short-lived. The unforgiving arithmetic and relentless pace of mass production quickly removed these robotic workers from the lines. Exorbitant costs and low efficiency demonstrated that humanoids in their current form cannot replace human workers in car manufacturing. Automakers are now drastically scaling back their ambitions, pivoting toward human-robot synergy and relegating humanoids to simpler tasks while exploring applications beyond factory work.
4. [para. 8][para. 9][para. 10] Traditional industrial robotic arms, pioneered by carmakers in the 1960s, remain mature and cost-effective. Zhou Jun, head of Geely's smart manufacturing department, noted that a heavy-duty industrial robot lifting 270 kilograms costs only a fraction of a humanoid and boasts an average time-between-failures of 80,000–100,000 hours. Their only significant downside is inflexibility—any vehicle design change requires costly, time-consuming production line adjustments, which is why final assembly, requiring intricate wire-harness threading and nuanced adjustments, still depends on human hands.
5. [para. 12][para. 13][para. 14] Humanoids were billed as the solution to this final automation frontier, able to walk, crouch, and execute complex general-purpose tasks. But assembly-line reality shattered these expectations. Factories producing 60 cars an hour give a workstation only a three-second window to complete a task—a pace existing humanoids cannot match. An executive who trialed humanoids noted that even simple tasks like picking and sorting parts were hampered by poor accuracy and slow movement, and their startup partner had to artificially speed up promotional videos to make the robots appear viable.
6. [para. 16][para. 17] Final-assembly complexity vastly exceeds current humanoid capabilities. Figure identified humanoid forearms deployed at the BMW plant as the primary mechanical failure point. XPeng spent over a year trying to integrate its IRON humanoid into factories before deeming it economically unviable. Chairman He Xiaopeng stated that the dexterous hands—the most expensive component—were too quickly destroyed by factory wear and tear.
7. [para. 18][para. 19] Environmental factors also impair performance. Li Peigen of the Chinese Academy of Engineering noted that welding sparks, moving conveyor shadows, and sheet-metal glares can blind humanoids reliant on vision-language-action models. Tesla, meanwhile, is pursuing automation for its Cybercab using a new "unboxed" manufacturing process instead of humanoids, breaking vehicles into modular pieces and pushing final-assembly automation to 78%—signaling that specialized machinery, not humanoids, is the path to efficiency.
8. [para. 21][para. 22] The initial surge came from the belief that autonomous driving and humanoids share technological DNA. With AI advancement, automakers felt they had first-mover advantages deploying this technology physically. But preliminary trials show the leap from smart cars to humanoids is monumentally difficult: vehicles operate on roads with established rules and vast training data, while bipedal robots operate in complex 3D spaces with scarce real-world data.
9. [para. 23][para. 24] The industry is plagued by bottlenecks—from algorithms and high-powered AI chips to hand durability—making most humanoids lab experiments rather than viable products. Public demonstrations showing robots folding clothes or picking up eggs are carefully choreographed with pre-written code. A single failed step—recognizing a cup, navigating a room, or executing a pour—ruins the entire task.
10. [para. 25][para. 26] High R&D costs are forcing retreats. XPeng originally planned to spend 10 billion yuan ($1.5 billion) over a decade but revised this to at least 50 billion yuan in November 2024. Li Auto, once branding itself as an embodied AI developer, reportedly downgraded its humanoid project to focus on self-driving technology.
11. [para. 28][para. 29][para. 30] Automakers are not abandoning humanoids entirely but are trading science-fiction fantasy for pragmatic business models. Industry insiders draw parallels to autonomous driving evolution: just as carmakers started with driver-assistance systems before full autonomy, robotics companies must find "stepping-stone" scenarios. Hu Haifeng, a robotics cofounder formerly at Huawei, said human-robot collaboration—robots handling mundane tasks while humans manage complex manipulation—lowers immediate technical hurdles.
12. [para. 31][para. 32][para. 33][para. 34] A paradigm shift is also occurring in factory design: Changan Auto believes facilities must be retrofitted for robots. Chen Chao, cofounder of a Suzhou-based industrial AI firm, suggested redesigning objects to robot-friendly shapes—like changing a circular knob to something basic machines can easily manipulate. As the industry waits for technology to mature—experts believe three to five years away—automakers are deploying robots in simpler, lower-stakes roles: Geely trials robots for parts sorting, while BYD and XPeng develop humanoids for retail showrooms.
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新