Posted this month! Ideal officially releases the unpictured NOA, covering all models and can be driven nationwide

Last month, Li Xiang, CEO of Li Auto, said in a public speech that based on the existing computing platform, supervised L3 level autonomous driving is 100% achievable, and unsupervised L4 level autonomous driving will definitely be achievable within three years.

In addition, Li Xiang also categorically stated that Li Auto will launch a nationwide map-free NOA (navigation assisted driving) in the third quarter of this year.

As soon as these remarks came out, doubts continued to be heard. With Li Auto’s current intelligent driving capabilities, it is a big challenge to achieve no map in the country in the third quarter. On the one hand, Li Auto has a late start in smart driving. On the other hand, Li Auto's move to recall laid-off employees last month has caused a lot of public opinion.

Leifeng.com previously quoted employees from Li Auto’s intelligent driving department as reporting that Li Auto’s recent personnel adjustments have had a great impact on the intelligent driving R&D and testing department. Development has almost come to a standstill, and testing is only carried out by a small number of low-cost external teams. assist.

As a result, Li Auto, which had not started large-scale layoffs for only one month, recalled some laid-off employees in key positions.

Perhaps eager to prove itself, Li Auto held the Li Auto Intelligent Driving Summer Conference tonight, putting all its bullets on the table.

No picture NOA, push this month

At tonight's press conference, Li Auto officially brought a major upgrade of OTA 6.0 intelligent driving without pictures NOA. This upgrade will be completed within the month, covering MEGA, L9, L8, L7, and L6, with a total of 240,000 There are many ideal AD Max users, including the first batch of ideal L9 owners in 2022.

On May 10 this year, Li Auto recruited 1,000 users and officially launched the public beta of NOA, also known as AD Max 3.0. "User feedback is very good, far beyond our imagination." Fan Haoyu, senior vice president of Li Auto's product department, said. Therefore, Li Auto quickly expanded the push scale from 1,000 to 10,000 people on June 17, followed by Unlimited recruitment started on June 28.

As of now, Li Auto has more than 810,000 smart driving users, with cumulative smart driving mileage exceeding 1.9 billion kilometers, and per capita mileage exceeding 2,300 kilometers. Fan Haoyu said that Li Auto NOA has become "the most important tool and partner in life" for Li Auto users.

This updated unpictured NOA has four major features.

First of all, it can be opened nationwide.

Last night, Li Auto collected user data across the country who used the picture-free NOA between 18:00 and 20:00, and based on their location, formed a picture like this:

In other words, Li Auto’s unpictured NOA truly has the ability to be driven “all over the country”, “whether in cities or towns,” Fan Haoyu added.

The reason for such performance is that Ideal's current mapless NOA has been comprehensively improved in terms of perception, control and overall system capabilities, thus improving the NOA's bypass capability – this is OTA 6.0 mapless NOA. The second major feature , some public beta users said that the current ideal can be called the "King of Detour".

Detour capability is an important indicator for evaluating a city's NOA. If the system has strong detour ability, it can solve problems cleverly; if the system has weak detour ability, it can only fall into endless waiting and then be drowned out by the horn of the car behind.

In Fan Haoyu’s view, the key to achieving “smooth detour” is to improve the system’s “space-time union” capability.

The first is "space".

When a car drives on the road, it has two directions, one is front and back, and the other is left and right. In the past, I needed to control the front and rear before moving left and right, and it was separated, which caused lag. Now I think about it in parallel.

Next is "time".

If the sequence continues, we can predict all the information in the future. After considering all the objects, the system will form an optimal path, including the control of our vehicle, not only merging out, but also the subsequent driving. Slippery, why? Because the timing has been carried out in a round-robin manner, the whole process is very smooth.

The third important capability of OTA 6.0 image-free NOA, which Ideal internally calls “passing the intersection easily ”, is actually the system’s over-the-horizon capability without prior information.

From an ideal point of view, the so-called "God's perspective" behind an intersection that is easy to pass includes not only the seamless surrounding environment spliced ​​with cameras, but also road information and navigation information.

"For urban NOA, if it cannot be very easy and fast to pass through intersections, users will not use it continuously," Fan Haoyu said, "But our public beta users use it 97% of the time every day, which shows that its The experience will be great."

This involves the fourth feature of OTA 6.0 image-free NOA – tacit understanding and peace of mind.

Fan Haoyu said that "tacit understanding" and "peace of mind" are two words often mentioned by Li Auto public beta users. "You will feel that when driving this car, it gives you a feeling of being a comrade and tacit understanding. Behind this In fact, it is a very important ability improvement, which we call “decimeter-level micro-management.”

In addition to the above four major features, Li Auto also emphasized the active safety brought about by the advancement of intelligent driving at the press conference. If tacit understanding and peace of mind are the feelings and experiences of a single user, then security is the result of a large number of users.

Li Auto stated that in the first half of this year, the accident rate of Li Auto’s human driving decreased by 30%; during this period, Li Auto’s AD Max active safety system prevented a total of 360,000 potential accidents for users, and the AEB false trigger rate was reduced to less than 300,000 kilometers. , "This means that users are basically unable to touch it." Fan Haoyu said.

Regarding potential threats on the road, Li Auto has created a "safety risk scenario library", which is arranged in order from low risk to high risk, and from low frequency to high frequency, and the risk scenarios are divided into 9 types.

"Basically all our friends will start with high frequency and high risk in the upper right corner, but our steps do not stop here." Fan Haoyu said that the OTA 5.0 launched last year can already handle "low frequency and high risk" and "medium frequency and medium risk" These two parts of the threat include being able to completely stop when encountering a stationary large truck at 120km/h at night, and making emergency avoidance at 130km/h through AES (automatic emergency steering).

The OTA 6.0 released today will improve the vehicle's capabilities in "high-frequency and low-risk" scenarios.

For example, when parking in a garage or in the basement of a shopping mall, there may be a pillar that we cannot detect. In this case, we can actively brake. It has no blind spots in all directions. It is really So that users will no longer be troubled.

However, it should be noted that this version of the NOA without pictures still requires the driver to pay attention to the surrounding road conditions at all times and take over in advance if necessary.

The “big move” in the second half of the year

In addition to the unpictured NOA released within the month, Li Auto also mentioned its “big move” for the second half of the year during the second half of the press conference.

In "Thinking, Fast and Slow", Nobel Prize winner Daniel Kahneman used System 1 and System 2 theories to explain the process of human decision-making and thinking. Li Auto believes that their "big move" in the second half of the year will be Kind of like this pattern.

It's easy to understand here. System 1 is where people make quick decisions based on their past experiences and habits. For example, to answer the question "How much does one plus one equal?" we don't need to think at all. System 2 is the thinking and reasoning ability, which is specially used to deal with complex problems or unknown scenarios that require thinking or reasoning to solve.

Jia Peng, head of Li Auto’s intelligent driving technology research and development, believes that when people drive, “system 1 is actually working”, and system 2 may only be used in 5% of cases.

Li Auto's "big trick" is to work like humans through the cooperation of System 1 and System 2. Let’s first talk about its “System 1”, E2E (end-to-end).

In fact, the version Li Auto plans to push this month is a full-scenario NOA software stack that does not use high-precision maps at all, and its performance can roughly correspond to the later version of Tesla FSD v11.

Judging from the architecture of AD Max 3.0, Ideal has incorporated “prediction” and “planning” into the same neural network model. Its next evolution direction is to connect it with the perception model to achieve lossless transmission of information and achieve global optimization. End-to-end" architecture.

The so-called end-to-end solution refers to the entire process of "perception-decision-control" within a unified system framework, and training is completed through deep learning methods instead of decomposing it in the traditional way. It is composed of multiple modules such as perception, positioning, path planning, and control. The upper module outputs the results and guides the lower module to run.

In human terms, it means "AI drives the car."

Traditional smart driving is like a student who listens carefully but is not very good at divergent thinking. He knows everything the teacher teaches. Once he encounters something that the teacher has not taught, it is hard to say. The end-to-end solution will integrate knowledge and draw inferences from one example.

This is the "System 1" of the ideal NOA, and its "System 2" is a VLM (Visual Language Model), which has some logical thinking capabilities and will verify the decisions of "System 1" in some complex situations. In the end Realize vehicle control.

In addition, Li Auto has also added a cloud "world model" for reinforcement learning and decision verification, making this "student" smarter as he learns. "It needs to know what's wrong," Jia Peng said.

The other side of smart driving is the smart cabin

The future of travel is probably like this: driverless, people don’t have to drive, but they have to work.

Last night, He Zongcheng, chief content officer of Aifaner, shared his experience in Moments – using Apple Vision Pro to polish the copywriting for the sugar factory's new product "Xinlixian" in an online ride-hailing service. In his view, this is not just work, but a preview of future life.

Maybe one day, online car-hailing services will start driving on their own.

In our imagination of the future, smart driving and smart cabins are always bound together. Since the car can drive itself, you always have to find something to do. Maybe it’s chatting with friends, or you can do it alone. Hold a book, fiddle with the handle, or directly turn on the "work mode" like He Zongcheng did.

In addition to these, Ideal has also prepared another possibility for you – Mind GPT.

At the Ideal Family Technology Day held by Ideal in June last year, Ideal Classmate No. 1 changed from a voice assistant to an AI virtual life. In Ideal words: it changed from a "classmate" to a "teacher."

The ideal student under OTA 5.0 has multi-modal perception capabilities. Simply put, its "ears, eyes and brain" capabilities have been greatly improved. With the help of multi-modal perception, it can identify "open this", "Open that" command. To be more specific, today's ideal classmates can judge whether the driver is giving instructions to the ideal classmates or talking to the co-pilot based on the angle of the driver's head turning to the right.

This is just one part of it. The ideal classmate can be called a "teacher" mainly because of the "Mind GPT" cognitive model that was added for the first time.

If you follow the Weibo of Li Xiang, CEO of Li Auto, you may find that for a period of time, his Weibo posts were all posted from the “Mind GPT iPhone client.”

Backed by Mind GPT, Ideal Classmate now has capabilities similar to generative AI such as ChatGPT and Midjourney. You can ask it any question you can think of, or ask it to draw. In many cases, it can give Satisfactory answer.

At present, it seems that the ideal classmates supported by Mind GPT can play a role in two scenarios: weekend family excursions and answering daily small questions. To a certain extent, you do not have to pick up your phone and focus on the road – even if it is turned on NOA, too.

Anyone with wheels is interested and welcome to communicate. Email: [email protected]

# Welcome to follow the official WeChat public account of Aifaner: Aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo