A year ago, in early summer in Beijing, I accompanied a delegation of British entrepreneurs and investors on an in-depth tour of the Robot World in Beijing’s Yizhuang district. The rapid deployment of robotic applications in China was an eye-opener for our guests from afar. In a bright, immaculate exhibition hall with simulated production lines, robots methodically performed intricate tasks: precisely sorting parts on an assembly line, accurately picking medicine boxes based on prescriptions in a mock pharmacy, and even expertly extracting espresso to craft a perfect latte. All of this painted a highly futuristic vision of “Smart Manufacturing in China”.
However, beyond the polished showrooms of Beijing, another very real scene—this one in Hangzhou—still lingers in my mind. On the rooftop of Unitree Robotics’ office building, employees had built a robot testing ground filled with various terrain obstacles. There, I saw the very same robot dogs and humanoid robots that nimbly vaulted over barriers in promotional videos struggling to navigate complex, real-world environments without an engineer holding a remote control. Compared to the flawless performances of humanoid robots in staged settings, this stark contrast with reality left me with a sense of the embodied intelligence industry’s awkwardness on the eve of full commercialization.
Fast forward to this year. In just one year, Unitree Robotics successfully went public, surrounded by a dazzling aura. Yet, at the recently concluded 2026 World Robot Conference in Beijing, Unitree’s founder, Wang Xingxing, poured cold water on an enthusiastic audience of practitioners and investors under the spotlight of the main forum. He bluntly pointed out that humanoid robots have not yet truly entered the stage of large-scale application, and their biggest shortcoming remains a lack of general-purpose capability. In his view, the true “ChatGPT moment” for embodied intelligence—when a robot can enter a completely unfamiliar home environment and autonomously complete about 80% of daily tasks using only voice commands—is still some distance away: 2 to 3 years in an ideal scenario, or perhaps 5 to 10 years if one is more pessimistic.
Wang’s remarks stand in sharp contrast to the numerous “social media trending topics” generated by two recent robotics conferences in Beijing. After the hype surrounding these events, I can’t help but wonder: how far are we really from the “ChatGPT moment” for embodied intelligence?
