How do Ideal car owners view "Ideal No Picture NOA", can the real level be compa

tech | 2024-08-02

On July 5th, Li Auto held a summer conference focused on intelligent driving, with the core messages summarized as follows:

1. The Li Auto NOA (Navigation Assisted Driving) will be fully rolled out within July, with the full-automatic AES (Automatic Emergency Steering) and comprehensive low-speed AEB (Automatic Emergency Braking) functions also to be updated and pushed;

2. A brand-new autonomous driving technology architecture.

In fact, I also participated in the public beta test of the Li Auto AD Max 6.0.0 Beta version at the end of June, which allowed me to experience a part of the content of the Li Auto 2024 Intelligent Driving Conference ahead of time. After watching the conference, I would like to discuss from the perspective of a car owner how to view the actual level of Li Auto's NOA without maps, and what details I believe can still be optimized.

Advertisement

01

NOA without maps essentially covers urban driving scenarios

According to Li Auto, the NOA without maps that will be fully rolled out has four major capabilities improved—

1. It can drive anywhere, no longer relying on excessive "prior information";2. Smooth detour maneuvers, combined spatiotemporal coordination;

3. Effortlessly navigate complex intersections with a "God's eye view";

4. Enhanced AES and low-speed AEB, ensuring safety in special circumstances.

These four points can be roughly summarized as the core content of the first half of this press conference. In layman's terms, after the update of the NOA without maps, it has essentially resolved the issue of urban NOA coverage, making the user experience smoother and, through the upgrade of AES and low-speed AEB, enhancing safety in special scenarios such as high-speed driving and parking. For instance, fully automatic AES is activated when AEB intervenes but cannot stop in time, and it avoids obstacles by making emergency steering maneuvers.

In other words, this update of the Ideal NOA without maps has essentially solved the problem of full coverage in urban areas and highways for NOA, and the user experience can be compared to the evolution from the earliest high-speed NOA to the current mainstream high-speed NOA in terms of usability. The main focus is on achieving theoretical and, in some cases, faster and safer judgment and operation capabilities than human drivers when facing complex scenarios.

However, looking at the entire press conference, the current Ideal AD Max 6.0.0 version seems not to have fully integrated the urban NOA and parking scenarios (the core content of the second half). You could also understand that it can help alleviate driving fatigue or assist in parking scenarios, but the transition between the two scenarios still requires human intervention, which essentially matches most of the features of Huawei ADS 2.0 (after the July update).

Finally, speaking from the perspective of a car owner, here is a brief discussion on the current experience of the AD Max 6.0.0 public beta version:

1. The driving strategy has become a bit more aggressive, primarily in terms of following distance and braking timing. The former, after adjusting the following distance to the first gear, will leave only about one parking space between the vehicle and the one in front during congested traffic, reducing the worry of cars cutting in line. The latter, in medium to high-speed driving conditions, applies the brakes later and with greater force than I would, but still ensures a safe distance from the vehicle ahead. Secondly, in terms of lane changes, it will attempt to change lanes whenever there is an opportunity, although occasionally it may also result in driving over the solid line or missing a turn.2. When making a right turn or navigating a curved road on an elevated structure or in an urban area, there is a habitual tendency to initially steer to the left before proceeding, regardless of the degree of the turn. To be honest, it felt a bit unnerving the first few times.

3. The UI interface has undergone detailed optimizations, including in graphics, color, and breadth, making the visual experience more technologically sophisticated, more refined, and closer to real road conditions.

4. The AD Max 6.0.0 version is likely not immediately available after an OTA update. After I updated to the beta version, the update notes included one condition for full-scene NOA activation: "Intelligent driving mileage > 1000km." Personally, I believe such a setting is reasonable. If one has not yet had sufficient experience with relatively less complex intelligent driving like highway NOA, directly using urban NOA without maps could lead to unexpected situations when encountering special circumstances.

02

Dual Systems, Next-Generation Intelligent Driving Solution

The core content of the second half of the press conference, "End-to-End + Drive VLM Solution," will be the focus of Li Auto's upcoming efforts.

Comprised of vehicle-side System 1, System 2, and a cloud-based model, System 1 is the end-to-end model (E2E) designed for rapid response to common driving issues. System 2 is the VLM visual language model (with thinking and reasoning capabilities, such as reminding to switch navigation routes when missing a turn or taking a wrong path in urban NOA without maps). Finally, a world model is used to verify the judgments and capabilities of System 1 and System 2, ensuring the overall system's safety and reliability.

In other words, you can understand that the next-generation solution is based on a BEV end-to-end large model responsible for the main process of assisted driving, with an additional DriveVLM model to detect objects affecting intelligent driving or complex intersections, providing a safety net for the former and addressing generalization issues. The two systems handle 95% and 5% of the functions, respectively, with System 2 primarily serving as an auxiliary to System 1.The purpose of the dual system is clearly to further implement the No-Map NOA function and other functions (such as parking scenarios), similar to System 2 to assist in handling complex scenarios like entering parking lots and parking spaces, in order to unleash the full capabilities and potential of the No-Map NOA.

03

Jia Shi Summary

Looking at the current level of Li's No-Map NOA, Li has basically caught up with most of the functions and standards of Huawei's ADS 2.0 smart driving solution, and the iteration route of the next-generation smart driving solution, One Model, is also similar to the GOD + PDP dual-network of Huawei's ADS 3.0. The core difference actually lies in the different judgments and handling of detail issues by the two companies.

From a personal usage habit perspective, after the No-Map NOA was pushed to update, it basically solved most of my needs for smart driving scenarios. However, there are occasional instances where NOA, due to coverage issues, such as when NOA is launched or switched to LCC, the vehicle will suddenly turn the steering wheel left and right, and the uncertain driving route is the situation I like the least when using the NOA function. Especially when approaching toll booths or exit ramps, although NOA has a reminder that it will exit, the detailed handling after exiting is indeed not smooth enough.