Wang Naiyan's team consists of over a dozen core R&D personnel, including several former colleagues from his time at TuSimple.
The project distinguishes itself by applying L3 autonomous driving technical logic to embodied AI. Specifically, the team uses reinforcement learning during the pre-training phase of action models.
The methodology involves domain randomization to shuffle data and conditions. The model is guided by a few fundamental first-principle reward functions, such as physical laws and safety baselines, allowing it to learn through thousands of iterations of self-play and trial and error to improve generalization.
This move follows a trend of automotive executives entering the robotics sector. Other examples include former Xpeng autonomous driving head Li Liyun and former Xpeng VP Chen Yonghai joining Zhongqing Robotics, and former Li Auto AI Chief Scientist Chen Wei founding Xieyue Intelligence to create consumer-grade embodied AI for home scenarios.
