Beijing Colorful City is a shopping center located outside the Fifth Ring Road in Haidian District.

A friend in Beijing told me that this is a big rural sector, although the land price is not cheap.

The shopping center opens at ten o'clock every morning, and there is an attached parking building. Everything is so plain and boring. Unusually, there will be several mysterious signs from internet giants in the building, and several mysterious camouflage cars parked around the second and third floors.

You can't miss it, because these cars are not only ostentatious enough in disguise, but also slower than ordinary cars, as if they have their own upgraded shots, emphasizing the epic sense of existence in the middle of the road. When you meet some impatient teachers following you, it will be lively. No matter how you honk the horn, your partner keeps quiet in the noise.

Don't do too much paving-the car is self-driving. Reasonable, I shouldn't say that the security officer didn't hold the steering wheel, or Robin would have to pay a fine again. But technically, people are really out of control, and the car is moving by itself, moving steadily.

Camouflage car is a brand-new SUV model that Weimar will release in the first half of next year. It should be the EVOLVE that appeared at the Shanghai Auto Show last year. According to the previous forecast, the production version of Concept will be equipped with L4 and 5G technologies. This L4 mainly refers to the L4-level autonomous parking function, which was jointly developed by Baidu and Weimar. The engineer told me that Baidu mainly provides technical capabilities, Weimar provides vehicles, and both parties define the scene.

At the Baidu World Congress held last month, this technology was actually displayed, and its scientific name was AVP. We also have an article ("Tesla of L5 will play with you, Weimar of L4 will play with you"). At that time, everyone was watching the cloud report live.

Not long ago, I was lucky enough to go to Beijing, which was first enlightened by L4 automatic parking, and had a very, very small-scale test ride experience. The feeling that entities are automatic is still great.

There are two links in the experience, namely, two scenarios that AVP currently plans to land:

One is HAVP with fixed parking spaces, such as your home, company or any other place where you hang out every day. You only need to take the bus once, and then the bus can reproduce the route on its own. This function will be the first to be launched when Weimar's new car is mass-produced next year, not futures, but at hand.

The training process is extremely simple, and the conditioned reflex is formed once:

For the first time, you just need to get ready and drive to the place where you get off every day, enter the on-board computer program to let the car confirm this starting point, and then you manually drive to your fixed parking space at a speed below 15km/h, which is actually to sweep out a route environment map with the car. After parking, there will be a prompt on the screen to start learning. The car will upload the map data to the cloud for analysis, and then download it back to the local area after learning, and you will gain skills.

It sounds like changing hands, but it's actually very fast, and it can be completed in a few minutes. You can also study locally. Some underground parking lots have no internet, so you can't upload data. You have to rely on local limited computing power to go shirtless, but the study time will be longer. According to the engineer's estimate, it may take more than ten minutes.

Similarly, the car can automatically park for you, but it needs retraining. Even if the boarding and alighting points are just reversed and the lines are in the same direction but different, it needs to be taught again. The key is to go from a fixed point to a fixed point. The car will only go to the points you have taught according to the map, and will not move with you in real time. From this point of view, this is a highly technical application based on scene presupposition.

You may think it doesn't matter, because there are many similar functions on the market, such as reversing tracking, remote parking, automatic parking assistance and so on. Parking this project, looks like a tacky thing, L4 can't save it. But the details determine success or failure, and at least two points make this set of HAVP more worth seeing:

One is the low driving threshold. You don't have to be a perfect driver in training, because Weimar is not a four-wheeled Turing machine, trying to reproduce human driving habits and trajectories. It is to obtain a map, and then complete driving in a centered, low-speed, humble and safe way according to your own preset rules.

The second is that you can really leave the car. The goal of most parking functions in the market is actually to solve the problem of difficult parking posture. Whether it is to help you let go or get off the bus, at least it needs close observation and monitoring. For example, even if you get off the bus remotely, you will still be asked to stay near the vehicle with your car key. If you leave too far, the system will automatically quit.

And HAVP is to solve your "last mile" walking problem. Of course, the learning route for the first landing is definitely far less than "one kilometer". Considering the cost, regulations and safety, it won't be long, and it won't be long. At present, Weimar is preset at about 50 meters, but it is said that it may be adjusted when it goes on the market. According to the engineer, there is no problem with the three-digit ability at present, and it seems to be the same from the experience of the day.

Another PAVP will be even cooler, that is, in a parking building like Colorful City or other parking lots, it can find an empty parking space and lock it after you get off the bus, without teaching, and it can also guerrilla in real time, especially supporting automatic crossing up and down stairs.

It and HAVP are two systems, and many codes are interlinked, but the route is not fixed and the environment is more complicated, which is reflected in the technical difference, that is, high-precision maps are needed. This Baidu will be responsible for the opening of the mall, and it is good to update the high-precision map library through OTA.

In terms of single capacity, this is definitely more advanced, but it is still highly dependent on the scene-the design has a fixed getting on and off point, usually at the entrance of the elevator hall. When you get off the bus and run, the car will identify the floor by the sign, and then find your own parking space; When you want to use the car, you always have to come out of the elevator hall. Call the car with the mobile APP and just wait at the boarding point.

The speed is designed according to the usual speed limit of the parking lot. Go straight at 5-7km/h, and turn more cautiously at 3-4 km/h. The engineer said that the technical ability can be accelerated, and the front hardware can basically reach the level of 15km/h, which will be slightly close to the actual flow rate of the parking lot. Now people and machines are mixed together, and machines are often excluded.

Baidu once thought about contracting the whole building, but later lost the meaning of going into the folk and dancing with people, thus more clearly recognizing the wolf nature in real traffic-although this will not change the principle of the machine, so autonomous driving is easy to fall into Asimov's dilemma:

If two cars look at a parking space at the same time, the machine will take the initiative to avoid it, and people's reaction is that I will say goodbye to see who can't stand the embarrassment first. In densely populated business districts, the efficiency of finding parking spaces by machines may become lower. Well, you're free anyway.

At present, this function only has vertical parking, and then there will be oblique and lateral parking, and of course it depends on OTA.

This set of AVP is the first product applied on Baidu's autopilot computing platform ACU. In fact, the platform was officially launched at the end of July this year, with an estimated annual production capacity of 200,000 sets.

In this scheme of AVP, five cameras and 12 ultrasonic radar are used, and the interface between millimeter wave radar and laser radar is reserved to prepare for more functions in the future. However, it is said that Weimar's millimeter-wave radar, originally used for ACC, is now connected to the system by engineers from both sides for redundant design, which is definitely not safe.

According to the draft of Baidu and Weimar, this is the only "soft and hard integration, AI production platform" in China. It's awesome, but starting from the already crowded parking track, there is a feeling that the selected son is at the grassroots level. In addition to the shape-like function mentioned above, the most similar to AVP is Tesla's call mode, which is reflected by the industry and is a novice village task of autonomous driving.

First, low speed is environmentally friendly. The higher the speed, the higher the requirement for recognition speed, and naturally the higher the requirement for computing power. Second, the scene is relatively fixed and closed, and the variables inside are relatively controllable. The above two points, whether technical ability or safety regulations, will be more suitable for landing, so that technology can continue to explore in reality. Including Tesla, the distance, range and speed of the route are also limited.

Third, this is a very cost-effective function point, which can greatly enhance the experience in high-frequency scenes.

Weimar has a set of data showing that when parking, more than 30% of the time of skilled old drivers is wasted in the process of finding parking spaces and parking, and novice drivers still have level problems, which takes longer; When picking up a car, most people looking for a car may need to spend 10 minutes or more. Even without data, when we look back on our daily life, we will know how much headache we have.

This pain point used to be the starting point for a number of parking apps to start businesses, but the collective failure at that time was that people's willingness to spend more time solving this just-needed problem was very low, and inefficient habits were much more stubborn than Internet entrepreneurs imagined. Direct integration in the car is the only solution of modern lazy society.

Therefore, the future functions will definitely be highly integrated from batch vertical to more comprehensive scene fusion. This is also the reason why this set of AVP looks quite capable in the current parking products. Weimar said that they will be the fastest brand to land and mass-produce L4 autonomous driving technology-maybe it needs some angles at present, after all, it will take time for mass production of PAVP. However, with the continuous iteration of ACU platform, more comprehensive L4 technology is expected to be realized.

Baidu engineers said that for Baidu, the key to technical competition, for example, compared with Tesla or other companies, is not whether the function can be realized, but the generalization ability of a function; In addition, it should be developed according to the functions of vehicle regulations.

At present, many car companies are in contact with Baidu, but in the end they will take the lead. On the one hand, they have capital relations, on the other hand, as a new power enterprise, they will be more open to new technical cooperation. This will also give Weimar a competitive advantage. Weimar, blessed by AVP next year, will become smarter than now.

Many years later, when we stand on the road of autonomous driving and look back, I wonder if anyone will remember that in Beijing Colorful City, so many cars took a violent road, which was extremely slow and brilliant. Think of it this way. Maybe the land price in this big rural area can go up by another 20 thousand.

This article comes from car home, the author of the car manufacturer, and does not represent car home's position.