Humanoid robot startups are raising huge rounds of funding, and the selling point is well known. Human-shaped machines will soon find their way into warehouses and factories and eventually into our homes. However, the Wall Street Journal reports a more cautious view within the industry. Even the companies developing these bots believe expectations have gone beyond what works.
Executives and engineers say today’s humanoids are still too unreliable for the messy, multi-step work people imagine at home, even if they can handle simpler tasks in controlled environments. The goal is not just a humanoid, but a humanoid that does useful work. Humanoid robot company Agility Robotics has hundreds of Digit robots at customers like Amazon and auto parts maker Schaeffler, where they move items around a warehouse.
Useful work is still tight
This narrowness was expressed at the Humanoids Summit in Mountain View, where the founders tried to take the message back. Humanoids are not yet a clearly defined product and are a big idea that emerged before the market and technology were ready.
Some companies find favor early on anyway. Weave’s laundry folding robots are being used in some laundromats in San Francisco, while Persona AI is building a welding robot for a shipbuilding customer.
The security costs dwarf the hardware
Installation costs are the main reason why companies avoid using robots. The report estimates that for every $100 spent on deployment today, about $20 goes to the robot and the rest goes to devices and systems designed to keep people safe.
Humanoids could reduce guarding somewhat since they are smaller and slower than heavy industrial weapons. The report assumes Tesla’s Optimus weighs about 5 feet 8 inches and 125 pounds, and Unitree’s G1 weighs about 4 feet and 77 pounds. The leap from homework videos to powerful home computers is still huge.
The home robot is later
Public forecasts remain aggressive, including Elon Musk predicting “insatiable” demand and targeting one million Optimus robots per year by 2030.
Consultants also point to bottlenecks like training data as teams use VR headsets to teach robots and 3D models to speed up the process. Right now, the best signal is intentionally boring, deployments that run every day with real customers, plus clear task boundaries and transparent installation costs.
