Yushu Technology promotes the self-evolution of physical AI models, and Wang Xingxing reveals the latest pre-research direction. Wang Xingxing, founder and chairman of Yushu Technology, revealed at the main forum of the 2026 World Robot Conference that the company is promoting the self-evolution of physical AI robot models and exploring the automatic completion of robot control code development through large AI models.

Yushu Technology promotes the self-evolution of physical AI models, and Wang Xingxing reveals the latest pre-research direction. Wang Xingxing, founder and chairman of Yushu Technology, revealed at the main forum of the 2026 World Robot Conference that the company is promoting the self-evolution of physical AI robot models and exploring the automatic completion of robot control code development through large AI models. The day after Yushu Technology landed on the Science and Technology Innovation Board, Wang Xingxing said that the company plans to use the cutting-edge AI large model as the core, customize rules, experience frameworks and constraint tools, allowing the model to independently retrieve the latest academic papers, research results and open source solutions, automatically generate robot control code, and then verify it through model self-evaluation and manual review. Wang Xingxing pointed out that the improvement of basic model capabilities will promote the continuous iteration of the self-evolving system. This closed loop can also incorporate simulation data, real-world data and human behavior data into the training process. As the scale of robot deployment expands, test data and evaluation indicators continue to accumulate, which is expected to significantly improve the efficiency of robot development and iteration. At the conference, Yushu Technology demonstrated the evolution of robots from single intelligence to group intelligence, and from motion control to autonomous collaboration through multi-model collaboration. Multiple humanoid, quadruped and wheel-footed robots relied on the self-developed AI group control system to complete cluster collaboration, and the manned mecha GD01 was also unveiled simultaneously. Regarding the "ChatGPT moment" of embodied intelligence, Wang Xingxing believes that the biggest bottleneck is still insufficient generalization ability. When robots can complete about 80% of tasks in unfamiliar environments using only voice or language instructions, the industry may reach a critical point, which may take 2 to 3 years, or as slowly as 5 to 10 years. He believes that it is difficult for current robot models to effectively correct tactile errors in the last few centimeters and millimeters of delicate operations. The root cause is the deviation between the input and output of the AI ​​model and the real world. As technological breakthroughs continue, this type of cumulative error is expected to improve in the coming years. via AI News (author: AI Base)