Wang Naiyan, Xiaomi Auto’s reported L3 autonomous-driving technology lead, has left the company and is exploring a new physical-AI venture, according to CnEVPost. The departure is one of the clearest recent examples of senior Chinese autonomous-driving talent turning toward robotics and embodied AI.
The move has not been accompanied by a formal launch announcement from Wang or a named startup. CnEVPost, citing reports from 36Kr and 21jingji, said he has been meeting investors and aims to build foundation models for what the industry often calls robot brains. The outlet reported that the venture may use reinforcement learning during action-model pretraining, an approach intended to help systems generalize beyond the exact situations they encountered in training.
The facts matter because Wang comes from a part of China’s automotive technology industry where teams have already spent years working on perception, planning, simulation, and machine learning. The more open question is whether those capabilities can be transferred to machines that must grasp, walk, work, and adapt in far less constrained physical environments.
From TuSimple and Xiaomi to a new venture
CnEVPost said Wang joined Xiaomi Auto in May 2024, took responsibility for autonomous-driving technology and L3 work, and reported to autonomous-driving and robotics division general manager Ye Hangjun. Before that, he spent nine years at TuSimple’s China operations and served as the autonomous-trucking company’s China chief technology officer from 2019.
That background helps explain why embodied AI is attracting attention from automotive specialists. Autonomous driving requires systems to fuse sensor inputs, predict what might happen next, choose actions, and operate under safety constraints. Robots face related problems, but the physical setting changes. A car moves within a road environment and generally has a stable shape. A robot may need to manipulate unfamiliar objects, maintain balance, coordinate multiple joints, and respond to forces that vary from task to task.
The reported project’s focus on generalizable embodied models reflects that challenge. A system trained to perform one carefully defined action is not necessarily useful in a different factory, home, or service environment. The pursuit of more adaptable behavior is therefore central to the appeal of robot foundation models, though it is still a difficult technical and commercial problem.
Wang is not the only executive with an autonomous-driving background to move in that direction. CnEVPost noted that former leaders from XPeng and Li Auto have joined robotics companies or started ventures, while Nio is backing a physical-AI startup led by its autonomous-driving chief Ren Shaoqing even as he remains with the carmaker. The pattern has turned the automotive sector into a talent reservoir for China’s robotics push.
Why automotive talent is moving toward robots
The attraction is understandable. Autonomous driving teams have developed experience in machine learning, simulation, data collection, real-time software, and safety-oriented engineering. Those are valuable ingredients for physical AI. They also work in organizations that understand how to turn complex hardware-software systems into products, manage suppliers, and prepare technologies for deployment at scale.
Xiaomi itself illustrates the overlap. The company has invested in both electric vehicles and robotics while expanding its broader AI ambitions. Its factory-trained humanoid shown in Beijing is a reminder that the company’s interests extend beyond passenger vehicles. Wang’s reported departure therefore takes place against an internal backdrop where mobility, automation, and robots already intersect.
At the same time, the transition should not be overstated. Driving systems benefit from vast road data and can often test within structured simulation environments. Robot data are harder to collect, especially for dexterous tasks. Physical interaction also introduces challenges around force control, hardware wear, safety, and a much wider variety of objects and layouts.
The technical claims around the reported venture are still prospective. Reinforcement learning can help an agent learn from trial and error, but it requires careful choices about reward signals, environments, and safety. Pretraining an action model is not the same as proving that a robot can reliably complete commercial work in uncontrolled settings.
The competition for embodied-AI founders
The deeper significance of Wang’s move may be the competition for people who can bridge automotive AI and robotics. Investors have poured money into embodied-AI companies, but capital alone does not solve the shortage of leaders who understand machine learning and can direct complex hardware programs.
China’s startup landscape has already produced multiple approaches to robot intelligence. Some firms build complete humanoid systems; others focus on data, simulation, control software, or models that can work across different embodiments. University-founded teams drawing capital into embodied AI show another route into the sector, one based more heavily on academic research.
Wang’s reported effort belongs to a different but complementary group. It brings experience from autonomous driving, an industry that has had to confront real-time decision-making, large-scale data operations, and the discipline of shipping complex technology. If the venture is formally launched and funded, it will enter a crowded field with high expectations and many unanswered questions.
For Xiaomi, the departure creates a separate issue. Automakers are competing not only for sales but also for engineers who can build increasingly capable automated-driving and AI systems. Losing a senior technology leader can be disruptive even when teams are large. Across China’s auto industry, the movement of talent toward robotics may make it harder for vehicle companies to retain specialists just as software and AI are becoming more central to their products.
The story is still early. Wang’s venture has not disclosed a product, a legal name, financing, or a customer. What is clear is the direction of travel. The boundary between vehicle intelligence and robotics is becoming more porous, and one of Xiaomi Auto’s reported L3 leaders is now part of that migration.