EVENT DOSSIER
Wang Naiyan, the former technical leader of Xiaomi’s L3 autonomous driving technology, has now embarked on entrepreneurship in the field of embodied intelligence. The project is currently seeking financing. This entrepreneurial initiative aims to develop a embodied large model with high generalization capabilities, using its technical experience gained during Xiaomi’s research and development of L3-level autonomous driving. In terms of methodology, the project plans to introduce reinforcement learning in the pre-training phase of action models, rather than relying solely on human data for imitation. It advocates for “deductive” learning through domain randomization and extensive simulation trials, in order to overcome the current bottleneck of insufficient generalization in embodied intelligence. Although the industry believes that financing in the embodied intelligence field is becoming more attractive and that the leading players have a significant impact, Wang Naiyan’s team still faces challenges such as dimensional differences when transferring driving technology to robots, difficulties in data collection, and high concentration of funds.
原小米智能驾驶 L3 技术负责人王乃岩已投身具身大脑创业，目前项目正寻求融资。该创业项目旨在打造最“泛化”的具身大模型，复用其主导研发小米 L3 级自动驾驶时的技术逻辑，核心方法是在动作模型预训练环节引入强化学习，而非单纯模仿人类数据。业内认为当前具身智能瓶颈在于泛化能力不足，王乃岩主张通过域随机化与大量仿真试错实现“演绎”式学习，以突破纯模仿学习的局限。尽管具身智能赛道融资趋热且马太效应显著，但智驾技术迁移至机器人存在维度差异、数据采集难及资金高度集中等挑战，头部企业仍占据大部分资源。