How skilled and how intelligent robots are today
Two capabilities determine everyday utility: hand dexterity and physical AI. In terms of mechanics, robots are approaching human capabilities; the bottleneck is reliability.
Manufacturers count differently. Only Shadow is verified from primary sources; the rest are marked as targets or secondary.
Fixed gripper, pick-and-place
Multi-finger gripper, planned grips
Many joints, in-hand manipulation in demos, dynamic walking
Human-centric and reliable in everyday life, walking, standing up
Fine motor control and whole-body dynamics like humans
Sergey Levine (Physical Intelligence) sees fully autonomous household robots at the median around 2030, but warns that robust everyday robustness could take another up to ten years.
Rodney Brooks (MIT, iRobot) considers human-like finger dexterity unsolved for approximately 15 years. Bank of America calls the hand the largest cost block, almost 20 percent of a robot.
Source: Rodney Brooks, Why Todays Humanoids Wont Learn Dexterity
Physical AI: the brain
Robots no longer learn individually for each task; instead, they learn through large pre-trained models, similar to ChatGPT for language. The bottleneck is training data, because movement is not available on the internet.
A robot learns over 700 tasks
Web knowledge becomes action, 1 million movement data from 22 robots
First generalist robot foundation model
Foundation models for humanoids, generalization to new environments
Race for training data and reliability
Every movement hand-coded
One task per training
One robot, many tasks (RT-1)
VLA models, cross-embodiment (RT-2)
Generalization to new environments (pi-0.5, Gemini)
Generalist, reliable, economical
Mass production is already underway. State goal: Over 10,000 humanoids in commercial deployment by end of 2026, working rather than demonstrating. Unitree: 20,000 units delivered targeted for 2026, following roughly 5,500 in 2025. Around 85 percent of all humanoids delivered in 2025 came from China.
The best AI models and systems integration come from the USA and Europe. In real-world deployment, Western companies act more cautiously, deploying dozens rather than thousands of robots.
"The bottleneck is embodied AI, not hardware. We are one to three years away from the ChatGPT moment of robotics."
The point: China's lead is a manufacturing story, not an autonomy story. Even Chinese founders say full autonomy hasn't been solved. On the question of when robots will be fully autonomous, China and the West are close together.
Market data status: November 2026. Next update: December 2026. Prices come live, other figures are researched anew on a quarterly basis. Sources listed directly with each metric. No investment advice.