Focus hot discussion: analysis of Wuling's "memory driving", a common route for car owners to train themselves

Produced | Sohu Auto · E-Park

Author | Wang Wei

Editor | Cai Xinyu

(Data map)

In terms of grounding, Wuling said that he was second, and no one dared to be first.

As a result, the crazy pile of sensors can only achieve high-order auxiliary driving in a limited number of cities, which is obviously not what Wuling wants to see.

Therefore, Wuling teamed up with its old friend Dajiang to create a "high-order auxiliary driving" system that can be used in urban and rural areas without laser radar, relying on high-precision maps and memory driving at a small cost of about 5000 yuan.

——What is memory driving? Where is its innovation reflected?

——What problems does it solve?

——Why would I call it a "good idea"?

Today's article mainly answers the above three questions.

Before explaining the memory driving mode, please enjoy a passage from Dajiang Vehicle Demo video provided.

The highlight of memory driving is First remember the user's point-to-point driving route ， After saving the route, the system can complete point-to-point pilot assisted driving according to this route On the way, some traffic light recognition and automatic start and stop, obstacle avoidance and detour, close congestion and avoidance, courtesy to pedestrians, automatic acceleration and deceleration, automatic up and down ramp, intelligent speed regulation, paddle lane change, start and stop following vehicles in congested sections, etc.

I.e LCC with route memory.

Does it sound familiar? Actually Max version of memory parking, It just extends the memory route from the inside of the parking lot to the whole city.

In terms of process, memory driving is basically the same as memory parking.

The first step is that when this function is used for the first time on a route, the user needs to drive by himself and memorize the track, which is equivalent to taking a route for Ai drivers.

Step 2: After driving this route to the destination, click Save Route to let the Ai driver remember this route.

Step 3: When driving the same route again in the future, by memorizing the driving and selecting the previously saved memory driving route, Ai drivers can use the method you have taught, combined with real-time road condition data, to provide an experience similar to the Lite version of urban pilot driving assistance.

The memory driving process is not completely mechanical tracking, but also based on real-time road conditions to make decisions. For example, when you remember the route, you like to take the innermost lane. As a result, the construction will begin a few days later. At this time, the vehicles can avoid the road construction by themselves.

Automatically merge to the free lane.

Identify traffic lights, start and stop automatically, and complete unprotected left turn+avoidance.

Complete the collision avoidance of close range traffic jam independently.

Give way to pedestrians passing the zebra crossing.

Avoid motor vehicles and other traffic participants parked on the roadside.

Due to the fixed route, the system repeatedly learns and optimizes this track in the process of memorizing driving every day, constantly improving the driving experience and flexibly adapting to various driving styles.

It's like opening the fog of war a little bit. The more you open it, the smoother it becomes.

Obviously, "memory driving" is an affordable urban point-to-point automated driving solution to solve the daily commute problem of "one touch to work, one touch to home".

With this function, you can train a personalized automatic driving route to meet the most basic commuting needs, even if the manufacturer has not opened the city pilot auxiliary function in the city.

The birth of the "memory driving" function is inseparable from the limitations of the current high-level auxiliary driving.

Lidar is expensive?

Laser radar is not required for "memory driving".

In terms of external sensor foundation, "memory driving" can be achieved with at least seven cameras (one group of 8 million pixel inertial navigation front binocular camera, one 3/8 million pixel rear view camera, and four side view cameras). Compared with the common millimeter wave radar+ultrasonic radar+laser radar+7, 8, or even 9 cameras of new forces, The total cost can be controlled at about 5000 yuan, Cheap cars are also affordable.

The system also supports the expansion of additional sensors such as millimeter wave radar and laser radar to enhance the experience. The system is rich and frugal and can meet the functional definition and needs of different price models.

Is there a great demand for computing power?

"Memory driving" does not require excessive computing power.

In terms of computing power demand of domain controller, "memory driving" has a small demand for planning and control in the fixed route, so it does not need a large computing power chip. The minimum 32Tops of Dajiang self-developed intelligent driving domain controller can solve the problem, and the on-board computing power is about equal to the original Xiaopeng P7.

High precision map is not open?

"Memory driving" does not require high-precision maps at all.

Tianshi (market demand) has a good location (many cities have no high precision maps) and people (Wuling's ability to bring goods is blessed), and Wuling and Dajiang are smart enough to have both.

Through today's introduction, you will find that "memory driving" is a good idea in the business world.

In terms of cost performance, "memory driving" does not need any extra investment in technology and hardware, which can achieve a win-win situation for both brands and users.

As far as functionality is concerned, "Memory Driving" is aimed at the immediate needs of commuting to and from work. It aims to solve the pain points of user driving fatigue and add selling points to the product.

In terms of commercialization, "memory driving" is cheap and can be popularized to 100000 yuan A0 and small cars, so as to realize rapid optimization and iteration of models, algorithms and functions.

In terms of localization, "memory driving" can also expand many striking playing methods.

For example, "popular automatic driving routes to and from work are shared", one person's running map is shared throughout the city.

For outings, we can launch "popular travel routes sharing", where one person travels and everyone copies their homework.

Another example "Award for drawing" 。

Since the right to develop routes lies solely in the users themselves, they have a strong sense of achievement when participating in route construction. With a little encouragement from the brand, they can quickly accumulate data, train models, and optimize functions.

According to reliable information, Wuling, together with the "memory driving" function produced by Dajiang, will make its world debut on the Baojun Cloud model, which is also the third mass production vehicle that Wuling cooperates with Dajiang.

As a reference, the length, width and height of the cloud body are respectively 4295 × 1850 × 1652mm, the wheelbase is 2700mm, the maximum energy of the battery pack is 37.9 KWH and 50.6 KWH, the NEDC range is 360/460km, the maximum power of the motor is 100kW, and the all-around benchmarking price of BYD Dolphins is 116800 - 136800 yuan.

Without shortcomings, it is neither dialectic nor practical.

In fact, the trend of "memory driving" is not the first time that automatic driving (high-level assisted driving) has appeared for so many years.

Tusimple Tucson, an American technology start-up, has chosen the technical route of "point-to-point high-end assisted driving of popular freight routes in the United States" in the future, focusing on training the 48 busiest freight routes in the United States, and constantly improving the granularity of data, algorithms and models on fixed routes to achieve the business blueprint of "building a nationwide driverless freight network".

The "route patrol" function of Dajiang's own Jingwei series and Royal 2 industry customized UAV is also a practical application.

Both power grid and highway are fixed routes in essence, and UAV route patrol can not only improve efficiency, but also reduce casualties.

The defect of "memory driving" is actually hidden in the question.

First, the maps accumulated by "memory driving" are not complete maps, which can not completely replace the concept of high-precision maps, and can never achieve the accuracy and integrity of high-precision maps.

Second, because it requires repeated painting to get a better experience, the more times, the better the experience, which means that early users may have complaints about poor experience.

Third, due to the limited distance, it is only applicable to the company home, home school, company school, home parents' home and other high-frequency point to point commuting needs, which cannot reach the technical level and height of supporting high-level navigation assistance across provinces and cities.

[· Edit Summary]

When everyone exclaimed that only hundreds of thousands of luxury flagship models can realize the high-level auxiliary driving function, Dajiang beat down the price of intelligent driving and took the lead in realizing L2+auxiliary driving capability on the 100000 yuan A0 class car.

Even the high-level intelligent driving ability was delegated to the "Yueya" model at the level of 90000 yuan.

so to speak Breaking the inherent pattern of "intelligence only belongs to high-end models" 。

In addition, the concept of "memory driving" initiated by Dajiang and Wuling is being quickly learned and followed up by new forces.

Xiaopeng Automobile plans to launch the "commuting mode" based on XNGP in the third quarter of this year, claiming that it can provide users with targeted routes of commuting+high-frequency travel in any city.

From this perspective, People's Wuling has not only completed its own share, but also made additional contributions to users of other brands.

key word: