A screenshot from a Nio video shows the Nio ES8 using driver-assistance features.
- Nio said the update covers more than 700,000 users, with even those who bought their cars up to four years ago gaining access to the latest technology.
- The new version upgrades the driver-assist training framework to further enhance a more human-like driving experience.
Nio Inc (NYSE: NIO) began rolling out a new version of its Nio World Model (NWM) driver-assist software to vehicles on Thursday, with more than 700,000 cars set to be upgraded.
The company said even users who bought their cars up to four years ago can enjoy the latest and most advanced technology.
Nio said it is the first automaker in the industry to achieve co-developed driver-assist software across general-purpose chip platforms and in-house developed chip platforms, with synchronized release versions.
This means models powered by four Nvidia Orin-X chips and those using Nio’s in-house developed Shenji NX9031 chip can receive the update at the same time.
Nio said this is underpinned by the systems engineering capabilities it has accumulated since starting full-stack in-house development in 2020, as well as its forward-looking investments in hardware reserves and software capabilities.
At the chip level, Nio said it had earlier anticipated that memory bandwidth would become a key bottleneck for bringing large models into vehicles, and stocked up on computing power accordingly.
At the software level, Nio did not adopt the general-purpose toolchain provided by suppliers when using Orin-X, instead developing its own deployment framework and compiler.
Nio said its in-house developed AI compiler shortened the development cycle for new model operators from one to two weeks to one to two days, while boosting inference performance by 20%.
This update is another major upgrade following the version released in January this year. In the January version, Nio became the first in China to apply complete closed-loop reinforcement learning to driver-assist development.
In the new version, Nio upgraded its technical architecture to a complete three-layer training framework of “world model + supervised fine-tuning + closed-loop reinforcement learning.”
The newly added supervised fine-tuning training uses high-quality behavioral data to let the model learn human-like driving behavior, combining a high baseline with human-like and compliant characteristics.
Nio also said the new version is the first in China to enable driver-assist systems to directly output steering wheel and acceleration/braking pedal signals, rather than outputting sampled trajectories.
The company said this end-to-end kinematic modeling approach offers a shorter path and lower latency, substantially improving the smoothness and precision of vehicle control.
In terms of experience, Nio said the new version’s route-selection accuracy is at an industry-leading level, even without using high-definition maps or enhanced navigation maps.
In addition, Nio’s NWM became the first among automaker in-house developed systems to achieve real-time recognition and understanding of tidal lanes and variable-lane overhead signs.
Nio said the new version achieves industry-leading performance on the two conflicting metrics of false braking and risk intervention simultaneously, balancing peace of mind with efficiency.
The company said that since the January version was pushed out, user duration and usage rate of Nio’s driver-assist features have doubled.
Meanwhile, Onvo, Nio’s mass-market sub-brand, also began rolling out the Coconut+ 3.0.2 version, continuing the same technical paradigm.
Nio said that because the underlying software is all developed in-house, it can complete the migration of mainline features to Onvo models on the Shenji NX9031 platform in as little as two weeks.
Onvo also plans to push a major upgrade covering the smart cockpit and driver-assist features to all Orin-X versions of its models by the end of July.
Nio said the update aims to combine a forward-looking technology paradigm with in-house systems engineering capabilities to raise the usage rate of driver-assist among users.
There is no system other than Tesla’s FSD that can rank among the first tier, a Li Auto executive said.
