The capabilities of humanoid robots are improving rapidly. At Tesla's event in October 2024, Optimus served beverages, and at the Spring Festival Gala in February 2026, a Unitree robot performed a dance.
Furthermore, at the 2026 World Humanoid Robot Competition, a robot ran 100 meters in 8.86 seconds, recording a time that surpassed the human world record.
However, experts are maintaining a cautious outlook. Karol Hausman, co-founder of Physical Intelligence, stated that while robots excel at acrobatic movements, simple tasks such as picking up a can of cola remain extremely difficult.
Some demonstrations have utilized teleoperation. It was reported that even the beverage service by Tesla's Optimus involved human operators controlling the robot remotely because, while the hardware was sufficient, the software was not yet mature.
Real-world labor requires advanced manipulation skills to handle objects. Learning physical tasks such as packing boxes or driving nails is far more difficult than learning a dance.
Furthermore, a lack of versatility remains a challenge. While humans can learn and execute diverse tasks instantaneously, methods to achieve similar general-purpose utility in AI robot models have yet to be established. The ability to complete complex tasks over extended periods is also still under development.