.png?w=1555&h=1555)
[Autohome News] On January 28th, we learned from NIO's official sources that a major new version of the NIO World Model (NWM) has officially begun its rollout. The first batch will be pushed to over 460,000 vehicles equipped with the 'Banyan' intelligent system. Vehicles with the 'Cedar' and 'Cedar S' systems will also receive the update in the near future. In this major update, NIO has, for the first time in China, applied complete closed-loop reinforcement learning to intelligent driving assistance R&D. Utilizing the leading technical architecture of 'World Model + Closed-loop Reinforcement Learning', it has comprehensively modeled urban and highway Navigation-on-Pilot (NOP) assistance. Additionally, intelligent parking assistance and active safety features have seen further improvements in technical capabilities and user experience.
According to official information, the new NIO World Model NWM version focuses on enhancing the fundamental driving capabilities of its full-domain NOP assistance, aiming to provide users with a driving experience characterized by accurate lane selection, strong interaction with traffic, and stable vehicle control, balancing safety and efficiency. Specifically:
-
In lane selection: The new NWM version optimizes the ability to choose lanes based on navigation routes, improves capabilities for avoiding traffic flow and making dynamic lane choices, and adds new efficient lane change features. Capabilities for reasonable lane selection at consecutive intersections, intersection navigation, and overall traffic efficiency on roads have been significantly improved.
-
In traffic interaction: The new version focuses on improving lane-changing capabilities. It can actively identify gaps in traffic flow to execute lane changes, making lane changes more decisive and smoother. It also maintains excellent performance in narrow roads, efficiently finding gaps to pass through.
-
In vehicle control: The new version achieves smooth, integrated lateral and longitudinal control, with softer acceleration/deceleration and more coordinated steering and speed control. Additionally, the urban NOP assistance gains two new functions: intelligent speed control and human-machine co-driving. The former no longer rigidly follows the maximum cruise speed but combines road conditions and traffic flow for floating or defensive speed control. The latter can follow the direction of user intervention to deduce a reasonable driving path, appropriately controlling speed and direction.
Furthermore, the new NWM version introduces an Urban NOP Battery Swap function. This is the industry's first intelligent driving assistance function covering urban energy replenishment scenarios. It supports over 2,000 second-generation and above urban battery swap stations nationwide, seamlessly integrating with the Urban NOP assistance.
NIO is pioneering the R&D paradigm of 'World Model + Closed-loop Reinforcement Learning'. By conducting long-sequence reasoning within the world model and performing high-frequency closed-loop reinforcement learning, the model learns to understand the environment, grasp driving common sense, and incorporate human driving experience, achieving self-calibration in planning and execution. Under this paradigm, model training no longer relies heavily on expensive expert data. Instead, it can efficiently utilize vast amounts of human driving data and, through feedback mechanisms, develop the model's common sense and error-correction capabilities.
According to the official introduction, NIO's Banyan system vehicles have a computing power of 1016 TOPS. This major NWM update is being pushed first to over 460,000 Banyan vehicles, allowing users who purchased cars four years ago to enjoy the latest and most advanced technological achievements. Vehicles equipped with the Cedar and Cedar S systems, which feature NIO's self-developed, car-grade 5nm high-performance intelligent driving chip, the NX9031, will receive the update in the near future. (Compiled by Autohome, Zhou Yi)