NIO Autonomous Driving Unit Restructures, Multiple Top Executives Leave

NIO will roll out iterative versions of World Model 2.0 across multiple platforms from the end of this year through the first quarter of next year.

ChinaEVHome has learned that Bai Yuli, head of NIO’s artificial intelligence platform, and Ma Ningning, head of NIO’s world model, have recently left the company. Additionally, public information indicates that Huang Xin, NIO’s intelligent driving product lead, has also departed.

In response to these personnel changes, NIO stated that it has proactively reorganized its intelligent driving department to enhance the absorption of the latest general artificial intelligence technologies and to enable the efficient, rapid, and high-quality delivery of the intelligent driving experience.

The company explained that by establishing a “4×100 relay” model—comprising pre-research, mass production, platform replication, and model replication—it has merged the intelligent driving organization with the general AI team. This restructuring aims to create an organizational framework aligned with the development trends of general AI technology.

NIO added that the adjustment will better position the company to fully accelerate the development and delivery of World Model 2.0.

In fact, personnel adjustments within NIO’s intelligent driving department have been ongoing for some time.

According to incomplete public records, Yuan Hongyuan, former head of NIO’s spatiotemporal information department, left in November last year. Subsequently, following a second organizational reshuffle at the end of last year, Qin Hailong, former head of the deployment solutions and architecture department, also departed. In April this year, Fan Haoyang, an expert in planning and control, also left the company.

The simultaneous departure of these three core executives marks the most significant personnel shake-up to date within NIO’s intelligent driving division.

According to the plan, NIO will roll out iterative versions of World Model 2.0 across multiple platforms from the end of this year through the first quarter of next year.

The new version aims to build an open-set intelligent engine enabling open interaction by integrating linguistic capabilities with the world model’s spatiotemporal cognitive abilities. Users will no longer be confined to a limited set of commands but will be able to express themselves naturally, with the system correctly understanding and executing their intentions. Additionally, the new version will further enhance long-sequence capabilities through reinforcement learning.