Baidu automatic driving latest progress: unison in the next generation, June mass production, Robotaxi service will usher in industry big coffee

On April 29, Baidu Apollo announced the normal operation of opening a completely unmanned taxi in Beijing:

From May 2, Beijing citizens can experience completely unmanned Apollo Park through Apollo GoApp – “There is no driver, there is no safety officer”.

Baidu’s initiative will be the first to open a fully unmanned driving service to the public in China.

Baidu’s unmanned vehicle opens the normalized passenger service, which means that Apollo Go Robotaxi enters a new development stage.

On the same day on the 29th, the heart of the car took the lead in the Shougang Park experienced a trusted driver.

On the same day, we also have visited Baidu Vice President, and Wang Yunpeng, General Manager of the Automatic Driving Technology Department.

At this point in time, it is just nearly one year away from Baidu to the public. It is just nearly one year. Recently, Apollo’s automatic driving mileage has also announced 10 million kilometers.

In this interview, Wang Yunpeng revealed our exclusive:

After accumulating the 2 million kilometers of test, Baidu Apollo is new in Robotaxi technology.

Baidu Apollo fifth generation shared no car will fastest in June, this model has been identified with 2 car enterprises, and will be equipped with a customized version of laser radar. Alternately, a new generation of automatic driving systems will achieve “10 times higher than the upper performance ratio of the performance ratio”.

Based on the new generation of automatic driving system, Baidu Apollo Robotaxi will implement technical solutions with the model of the front load mass production, and truly realize the data of the mass production vehicle and Robotaxi.

Based on the fifth generation model, Baidu will put 3,000 shared no-car in 30 cities in the next three years. It is expected that the number of operating vehicles in the end of this year will reach 500 levels.

Baidu will also speed up the open operation and commercialization of Robotaxi. This section will soon usher in a business big coffee to join.

First, the fifth generation shared unmanned car is about to launch: equipped with custom laser radar, realizing the data coexis of the front mass production and Robotaxi

At present, the main model of the Apollo shared drone is the fourth generation, that is, the red flag EV Robotaxi with an FAW red flag.

This is also the fourth-generation model in the first steel garden.

The roof of the fourth-generation model is equipped with a 40-wire laser radar of a Herpet, integrated with 9 cameras, and millimeter wave radar and ultrasonic radar.

This program is different from Baidu’s existing mass production, passenger, passenger car solution, Apollo ANP.

The industry is generally believed that the more similar the sensing hardware architectures mounted in the vehicle, the higher the data acquired by the road test operation.

Under such conditions, the large-scale front mass production of the vehicle will greatly accelerate the technology advancement of Robotaxi.

Wang Yunpeng revealed that Apollo’s fifth-generation car will happen “changes in the naked eye.”

From the fifth-generation car, Baidu Robotaxi will use a basic-consistent sensor scheme with the mass production passenger car ANP, including 12 cameras, 6 mm radodes, and 12 ultrasonic radar.

In addition, Baidu Apollo has customized 2 laser radars with partners, 1 for mechanical rotary laser radar, 1 is mixed solid laser radar, in the Apollo ANP scheme of Robotaxi schemes and upgrades.

The customized version of the laser radar has five major features, some of which include:

Tight combination of hardware and perceived algorithm, improve accuracy and effective lift;

Underlative, increasing hardware failure and safety self-testing, such as severe influence performance such as stain adhesion, etc. The sensor will issue a alarm to the system.

The fifth generation model of custom laser radar will be tested in June this year.

Through this upgrade, Baidu Apollo will compress Robotaxi’s “half of the cost, and 10 times performance.”

Second, 3 years, 30 cities, 3,000 cars, Apollo Robotaxi will usher in commercial operations big coffee

In December 2020, Baidu proposed a plan to deploy 3,000 cars for 3 years in Apollo Robotaxi.

This time, Baidu also reaffirmed this plan.

Wang Yunpeng shares the large-scale commercial deployment of Robotaxi, the core is: technology, product, operation.

Technically, the AI ​​driver continues to be polished.

Each generation of automatic driving system is 10 times, this 10 times has many specific metrics:

Initially, a simple MPI (Miles Per Intervertion frequency), which gradually evolved to the success rate of regional points to the point delivery, the passenger’s ride experience.

The ride experience can be further refined, such as waiting for the length of the order, driving in a sudden buffer setback rate, etc.; reliability is also an important measurement indicator, such as in the past, some parts are scrapped less than a year, now I can use 3 – 5 years. On the product, Robotaxi’s core products are automatic driving pre-load bunters.

In Baidu, the deployment plan of 3,000 Robotaxi, the fifth generation model will be the main model of the launch.

This year, Baidu will put 300 fourth-generation models, and about 200 fifth generation models.

At present, the fifth generation drone, Baidu has locked two car factory partners, soon there will be a third addition. Three car companies in the above cooperation are domestic independent brands.

“Intelligent driving era, we believe that China’s car must be the best world.”

In operation, Wang Yunpeng said that it is not the end of the public, but a starting point.

From this point, you should accept the inspection of the masses and continue iteratively optimization. Safety is absolutely first, and the user experience cannot compromise. It is better and shorter than the human driver, and the pick-up time is shorter. More and more, gradually focus on the travel experience of the drone from an option into preferences and must.

Robotaxi wants to achieve commercial and profitability, to achieve a certain scale.

Rough measurement, at least 20% permeation rate in the local travel market.

If the car is too small, the capacity is insufficient, and users often can’t get the car without this service.

If there are too many cars, there is not good luck. It is necessary to increase efficiency, match the matching and dynamic balance of capacity supply and travel needs.

So regional selection, site selection and the size of the team deployment are key points.

Wang Yunpeng revealed that Baidu Apollo Robotaxi will soon usher in a big-trip to the big latch, this big coffee will lead Apollo Robotaxi’s travel service operation.

At present, through the opening operation of Changsha, Zhangzhou, Beijing, Guangzhou and other places, Baidu has established a complete set of operational deployment:

First, get a policy permit in the local area to ensure safety legal compliance;

Second, according to traffic big data, the analysis determines which regions are the most prosperous point of travel and set up a unmanned car stop station, covering the open road network and the region of travel demand; collecting and verifying high precision maps.

Third, key indicators such as standardized testing, improve safety, experience and delivery power;

Finally, after some key indicators reach the expected standard, open seed users perform multi-wheel experience optimization, and then enter a comprehensive open phase.

During this Shanghai Auto Show, Baidu showed the automatic driving open experience based on the ANP program in Jiading Anting.

Wang Yunpeng revealed that this experience, from the vehicle in place to the first customer test, only for less than 10 days.

Third, the automatic driving runs over 10 million kilometers, Baidu’s unique experience

Currently, only 2 automatic driving mileage accumulated by 2 companies have more than 10 million kilometers, the US Waymo and China’s Baidu.

China-US traffic is different, the United States travel is based on high-speed roads and fast roads, and the mileage accumulation is fast, and the driving scene is relatively simple.

And Baidu’s domestic test, 90% is a city-complex scene.

At present, the industry also has a mainstream view that the most difficult part is not to increase the test mileage after reaching a certain scale, but a valid difficult scene data is mining.

Facing thousands of kilometers of test data, Wang Yunpeng shares:

Baidu Apollo did data mining at the first stage of manual marking.

In the earliest period of test, Baidu has been targeted to the system’s deficiencies in the car.

The second phase is offline mining.

Since five years ago, the way offline data mining is started, that is, after the data is returned to the data center, the data is tagged by dynamic scene semantic understanding (such as the behavior of the vehicle, in carrying out no protection left), Analyze the typical scenes and problems, perform screening, weather, different types of obstacles, etc., and enter the simulation scenario.

The third phase is online excavation.

As the vehicle is increasing, the size of the data acquisition of the car is getting bigger and larger, so Baidu puts the data mining directly in the end, triggered by dynamic conditions, and collect high value data as needed, which can significantly reduce the storage cost of data. And the time and cost of later processing.

The road test mileage reaches 10 million kilometers, which is equivalent to the unmanned area of ​​the world’s automatic driving field.

Wang Yunpeng said that the company only achieves such road measurement levels, and it will form a cognition of how to make mine and alchemy.

From April 2020, open automatic driving taxi in Changsha is now nearly a year. Wang Yunpeng also mentioned that the real open operation has brought a lot of new awareness compared to his test.

For example, in the test phase, the core pays attention to security, and users will pay more attention to ride experience, especially insecure.

Baidu Apollo has done a special item in order to optimize passengers:

Using data mining to analyze how much the Robotaxi team is a hundred kilometers, and then compares the average of the human driver and the data of the national carrier driver. According to APOLLO statistics:

The human driver will have a hundred kilometers of driving, there will be 11.7 times can be stunned.

After professional training, the national team driver, from the Capital Airport to the country, a special brake that can be perceived by 100 km, probably less than 5 levels.

When Baidu Apollo Robotaxi started, for security priority, the brakes can be 3,40 times.

For user feedback, Baidu has established a special project for some physical optimization, and the target is better than the ordinary human driver.

After hard work, its Robotaxi team is now able to do a hundred kilometers less than 7 levels in the same area. Baidu’s next goal is to enjoy the ride experience of the national team.

From a user’s perspective, the user’s experience to Robotaxi is compared to the human driver.

Apollo Robotaxi’s AI old driver must be able to take a better ride than the human driver, while cost is lower.

“Unmanned driving system is depreciated in 5 years, if it is 350,000 yuan, it is 70,000 yuan per year, this level is lower than the car driver; if it can do 200,000 yuan, it is lower than the fast train driver.

With more convenient and very cheap travel services, users don’t need to buy so many private cars, the efficiency of the whole society will increase, the traffic congestion is significantly reduced, can reduce carbon emissions, China has achieved carbon peaks earlier, carbon Neutral, will result in huge social value. “

Wang Yunpeng believes that “this is very exciting, worthy of our struggle.”

Fourth, Baidu Apollo’s unique path in China: “Do not succeed with it.”

In the automatic driving area, Waymo and Tesla in overseas companies are two typical representatives.

Waymo’s route is ultimately unmanned, so Waymo has to build strong bicycle intelligence, enhance the perceived ability with expensive sensors to ensure safety and efficiency.

Tesla is a model for terminal consumers. In this direction, Tesla hopes to provide better human-machine experience, unmanned is not the purpose of Tesla.

Wang Yunpeng frank:

In the past few years, there is a long time for choosing what kind of route to choose, and it is gradually clarified – will not follow any one, it will not be successful, and it is necessary to combine its own resource endowment with China’s environment. Differential roads, use AI old driver + front load production vehicle + car road to co-+ 5G Yun Dynasty. “We believe that this can be achieved faster”.

Next, Baidu Apollo’s fifth-generation shared unmanned model will be an important node.

The pre-installation of the pre-installation of the pre-installation of the ancenta-to-laser radar, which provides a better automatic driving experience for more passenger cars.

The data of the mass production car is inbred, and the technology improvement of Robotaxi can be accelerated.

“We have chosen a way for Baidu, suitable for traveling in China. This route is neither Waymo route, nor Tesla route. But we believe in this road, we will go faster, you can achieve automatic driving earlier Large-scale commercialization. “

Leave a Comment

Your email address will not be published. Required fields are marked *