At the 2025 Tengchong Scientist Forum, Wang Xingxing, founder of Yushu Technology, gave an in-depth explanation of the "anthropomorphic" logic and practical challenges behind the development of humanoid robots.Wang Xingxing said: "People like robots that look like humans." The reality is that the better-looking and human-like the robots are, the more consumers are willing to pay for them.
He also mentioned that current AI technology is highly dependent on data-driven, and data quality directly determines the upper limit of AI capabilities. As far as humanoid robots are concerned, high-quality, large-scale action and behavior data are still scarce.
Wang Xingxing showed the participants the progress of Yushu Technology in this field:By collecting real-person movement data in advance and training in an AI environment, robots can reproduce the diverse movements of humans, realizing "the robots move as humans move."
Wang Xingxing also admitted that the current AI models in the field of embodied intelligence are still insufficient. "The current stage is similar to the state one to three years before the advent of ChatGPT - the direction is clear, but the breakthrough results have not yet been fully revealed."
He thinks,If a robot can complete about 80% of tasks through language or text instructions in 80% of unfamiliar scenes in daily life, it can be regarded as a "GPT moment" in the field of embodied intelligence.At present, robots perform well in a single task, but once the object or environment changes, especially when faced with the superposition of multiple tasks, the success rate will drop significantly, which reflects the robot's shortcomings in generalization and reasoning capabilities.
Finally, Wang Xingxing revealed: "Maybe I don't have to be present at next year's Tengchong Scientists Forum in person. I can send a robot to attend the meeting on my behalf. I can remotely control it from Hangzhou and let it synchronize my language and movements to achieve real-time interaction. This technology is already feasible and is expected to be put into practical application next year."
