Humanoid robots could follow verbal commands as early as 2027, Qianxun co-founder says
Gao Yang, assistant professor at Tsinghua University and co-founder and chief scientist of Qianxun Intelligence, told Reuters that humanoid robots could understand verbal commands and complete most general tasks as early as 2027, though it will still take time for them to enter households and take on practical work.
Gao said competition among Chinese humanoid robot makers is shifting from hardware to software, as companies aim to improve robots' intelligence so they can perform tasks with real economic value. This field is often called embodied intelligence. He described the 'brain' as the weakest link in the robot technology stack, and expects a 'GPT-3.0-level' milestone by mid-2027, when users can talk to robots in natural language and the robots will perform a series of reasonable physical actions to complete tasks.
The wider industry is waiting for its 'ChatGPT moment' — a software breakthrough that can turn frontier technology into practical products for a broad customer base. OpenAI's launch of ChatGPT in 2022 brought generative AI into the mass market and demonstrated its commercial potential.
To train its robots, Qianxun has deployed about 1,000 contract workers nationwide to collect real-world data. Currently, Qianxun's robots achieve a 90% success rate on simple tasks in structured living-room environments. Gao expects the next one to two years (2027-2028) to be the initial window for industrial applications, with robots entering commercial service scenarios around 2028 to perform relatively simple tasks. Entering households, he said, is much harder than industrial and commercial service settings.
When the company was founded, Gao said, its robots could only handle separate tasks such as pouring water or folding one piece of clothing. Now they can move across large spaces and carry out complex workflows continuously. Still, fine-grained actions like unscrewing bottle caps and dealing with previously unseen tasks remain difficult challenges for robots.
