Humanoid robots’ ChatGPT moment years away

Humanoid robots may be a decade away from their “ChatGPT moment,” Unitree Robotics founder Wang Xingxing has said, following his company’s soaring market debut.

Unitree debuted in Shanghai on Wednesday and soared to close 460% above its 150.80 yuan offer price, more than five times the price at which the company priced its $905 million initial public offering, amid huge interest in humanoid robotics as an emerging sector.

But speaking Thursday at the World Robot Conference in Beijing, Wang said a true breakthrough for robotics — its “ChatGPT moment” — will be when a robot can be placed in an unfamiliar home and complete about 80% of tasks through text or voice commands alone.

That could “arrive in two to three years if things move fast, or five to 10 years if they move slowly,” he said, according to CNBC’s translation of his speech in Mandarin.

The forecast is more cautious than Wang’s speech at the same conference last year, when he said such a moment could come in less than five years. The release of ChatGPT in November 2022 proved a game-changing moment for generative AI, helping ignite a bull run in tech stocks and the data-center buildout.

Wang’s remarks underscore the gap between investor enthusiasm for humanoid robots and their current commercial capabilities.

Unitree’s robots remain less efficient than human workers and must be retrained from scratch for each new task, Wang said. That inability to generalize across tasks and environments is the central obstacle facing the entire industry, Wang told the conference.

Unitree’s stock dropped nearly 19% to close at 687 yuan on Thursday, according to LSEG data.

Founded in 2016, the company rose to prominence after videos of its robots dancing and throwing kung fu kicks went viral. But whether the technology behind the spectacle can scale commercially, with AI and software constraints unresolved, remains the question hanging over the stock.

Wang said in Thursday’s speech that the core challenge is perfecting the precision of robots’ movements, as they can reliably plan broad movements but fail at “the last few centimeters or millimeters.”

To close the gap, Wang said Unitree was pursuing a “self-evolving development loop,” where AI models write and test the robots’ control code, and each round of results is scored and fed back in to sharpen the next version.