Why? These poor Western producers are losing competition to China. They have to explain their inability by mystical “difficulties” of robotics development. This helps also to put a layer of negative propaganda on affordable Chinese robots, so less people will demand to try them. Let’s underline that the West is not directly competing with the Chinese humanoids, since the Chinese are not making AI-based expensive humanoid butlers for the rich:
Western humanoids, butlers for the rich, since deindustrialized societies can’t produce anything affordable for everybody:
General-purpose intelligence for complex task reasoning and autonomous operation in human environments. Tesla Optimus learning from real-world data; Figure 02 using OpenAI models for task reasoning [1]; Boston Dynamics' Atlas performing agile, dynamic movements. Higher cost; focused on scaling premium technology (e.g., Tesla aiming for <$20K).
Chinese humanoids, cheaply produced in well-developed industrial settings, to do the work in factories, logistics and households:
Physical dexterity and speed, often targeting specific industrial or commercial applications. Unitree H1 performing dance routines; Xpeng's IRON achieving lifelike movement; TARS robot performing delicate two-handed embroidery. Significant cost advantage; rapid mass production (e.g., Unitree R1 priced under $6,000; China on track to make 10,000+ units in 2025).
“Billions of dollars are flowing into humanoid robot startups, as investors bet that the industry will soon put humanlike machines in warehouses, factories and living rooms.
Many leaders of those companies would like to temper those expectations. For all the recent advances in the field, humanoid robots, they say, have been overhyped and face daunting technical challenges before they move from science experiments to a replacement for human workers.
"We've been trying to figure out how do we not just make a humanoid robot, but also make a humanoid robot that does useful work," said Pras Velagapudi, chief technology officer at Agility Robotics.
Agility currently has hundreds of its Digit robots working with customers, including Amazon.com and auto-parts company Schaeffler. They perform tasks including picking up items and moving them around a warehouse.
As robots like Digit have begun to find niches of demand, some analysts and technology executives have begun to predict a looming humanoid robot wave.
Velagapudi is skeptical. Getting human-shaped robots into warehouses or industrial sites to move boxes is one thing, he said; building a robot butler is beyond the industry's current capabilities, with current robots too unreliable to perform complex tasks.
Then there is safety. According to a survey of executives, the cost of installing robots is the biggest reason companies avoid deploying robots, said Ani Kelkar, a partner at McKinsey. For every $100 spent on deploying robots today, only around $20 is on the actual machine, with the rest being spent on equipment and systems designed to protect humans from injury, Kelkar said.
In theory, a humanoid robot won't need the same safeguards as an industrial arm that might weigh thousands of pounds and operate at high speeds. Tesla's Optimus robot stands approximately 5 feet 8 inches tall and weighs 125 pounds; Unitree's G1 is even smaller at 4 feet and 77 pounds.
But the gulf between the promise of the technology and what it can do today is wide, Kelkar said. "We're doing a big extrapolation from watching videos of robots doing laundry to a butler in my house that can do everything," he said.
At the recent Humanoids Summit in Mountain View, Calif., touted as the world's largest gathering devoted to the subject, Isaac Qureshi used a virtual-reality headset to control an early prototype of a robot designed to clean office spaces.
The chief executive of the recently formed company that makes it, Gatlin Robotics, followed behind the robot as it attempted to scrub a brick wall.
"Slowly, we're going to teach the Gatlin robot more things, like starting with dusting, surface cleaning, trash bins and then the toilet," said Qureshi. "Toilet's a big North Star."
On stage at the summit, one startup founder after another sought to tamp down the hype around humanoid robots.
"There's a lot of great technological work happening, a lot of great talent working on these, but they are not yet well-defined products," said Kaan Dogrusoz, a former Apple engineer and CEO of Weave Robotics.
Today's humanoid robots are the right idea, but the technology isn't up to the premise, Dogrusoz said. He compared it to Apple's most infamous product failure, the Newton handheld computer.
Launched in the 1990s amid a wave of hype about personal digital assistants, the Newton was a commercial failure that was canceled after only a few years. Just a decade later, however, handheld computers would become ubiquitous with the launch of the iPhone.
"Full bipedal humanoids are the Newtons of our times," Dogrusoz said.
Weave is building laundry-folding robots, which are being used in some San Francisco laundromats. But even the founders whose robots are finding some market traction see risks in encouraging the notion that the technology has arrived.
"I think we have to have this sense of responsibility about the timelines we are talking about, the adoption timelines," said Nicolaus Radford, CEO of Persona AI, in a keynote address at the summit.
Company leaders say there is a narrow set of roles in which humanlike robots make sense today, including performing simple, repetitive tasks such as moving boxes.
Persona is building a welding robot for a shipbuilding company, a function Radford said is ripe for roboticization because the danger involved makes labor hard to find.
For something like robot butlers, the market is farther off, he said.
The cautious, if not downright gloomy, outlook by leaders and engineers of humanoid robot companies stands in contrast to forecasts made by some of the biggest names in technology.
Elon Musk predicts that demand for humanoid robots will be "insatiable" and has said that Tesla aims to produce one million of the company's Optimus robots a year by 2030. Nvidia CEO Jensen Huang has said he believes the world is on the cusp of making everything that moves robotic. "Humanoid robots, the technology that makes it possible is just around the corner," Huang said on a podcast in January.
Optimists like Musk and Huang see a confluence in trends behind their predictions. Billions of dollars are being spent on data centers to train the artificial-intelligence models that will power future robots. The aging of populations in many countries means there will be fewer workers as well as a growing cohort of elderly people in need of care. Governments also see robots as a way to win manufacturing jobs back from overseas.
Beyond the macroeconomic trends, improvements in battery and motor technology mean that robots are becoming more adept at mimicking human motion and can work for longer periods. Earlier this month, the CEO of one of the hottest robot startups, Figure AI, posted a video of the company's latest humanoid bot jogging in a manner eerily similar to a human.
Dozens of robot startups are attracting huge investments, with around $5 billion being invested in humanoid robots this year, said Kelkar of McKinsey.
FEV Consulting, which advises many robotics companies, predicts that by 2035 there will be around one million humanoid robots at work. Holding back the expansion is a lack of training data for robots, with many startups using humans wearing virtual-reality headsets to train robots. Others are experimenting with 3-D models of workspaces to speed up the process.
But no one is sure how much training is required before robots can move from folding shirts to doing multiple household chores, said Dominik Boemer, a manager at FEV.
"Humanoids perhaps solve some specific problems, but it might not be as big of a market in the near term as everyone thinks it is," said Jeff Mahler, chief technology officer at Ambi Robotics, which makes package-sorting robots.
Ultimately, there is a more fundamental question to answer: Do we even need a robot with arms and legs?
There are downsides to the human form: Robots that look like us are prone to tipping over and engineers struggle to create a mechanical version of the human hand. We rely on sensations from our skin to know how much pressure to apply, something robot builders struggle to replicate. Some engineers say the future isn't in replicating the human shape but in improving on it, with four hands instead of two, or suction grippers instead of fingers.
"My point of view is that we are sticking to the humanoid form too much," said Max Goncharov, the chief technology officer at RemBrain. "In the factories, it's all about efficiency, and efficiency means more specialized robots."
"I think humanoids will do a tiny layer of tasks in factories in the future," Goncharov said.” [2]
No butler for you. No robotics in the West.
1. How did Figure 02 use OpenAI models for task reasoning?
"Figure 02" refers to the second-generation humanoid robot by the company Figure AI. Its initial task reasoning capabilities, demonstrated in August 2024, involved a collaboration with OpenAI where cloud-based OpenAI models processed human speech and visual data to generate high-level plans and commands for the robot to execute.
Figure 02 Task Reasoning with OpenAI (August 2024)
The initial system used custom AI models trained in partnership with OpenAI to enable the Figure 02 robot to perform tasks based on human interaction and environmental feedback. The process was as follows:
Speech-to-Speech Communication: The robot used onboard microphones to capture human speech commands. OpenAI models processed this audio input, converted it to text, and generated a text response, which was then converted back into a spoken response by the robot.
Vision Language Model (VLM): Figure 02 utilized six onboard RGB cameras to perceive its environment. An onboard VLM enabled "fast common-sense visual reasoning" from this camera data, allowing the robot to understand its physical world, avoid obstacles, and coordinate hand-eye movements.
Task Execution: The reasoning and planning for complex tasks were primarily handled by the cloud-based OpenAI models, which then sent commands and a high-level behavioral intent to the robot's control system for execution. This allowed the robot to combine speech and visual reasoning to understand a command (e.g., "put that item away") and take the required action.
Shift to In-House Helix Model
It is important to note that Figure AI later transitioned away from general-purpose OpenAI models for core task execution, deciding to use its own in-house developed Helix VLA (Vision-Language-Action) model to enable more specific and efficient on-device reasoning and control (as of early 2025).
2. Even Companies Making Humanoid Robots Think Their Promise Has Become Overhyped. McLain, Sean. Wall Street Journal, Eastern edition; New York, N.Y.. 26 Dec 2025: B1.
Komentarų nėra:
Rašyti komentarą