Carrot run, what technology can destroy the driver's job?
Latest update time：2024-07-18
Reads：
▲
For more exciting content, please click on
the blue words
above
and follow us!
In the past two days, RoboTaxi (driverless taxi) has started a fierce battle: on the one hand, "RoboTaxi" has become a hit in Wuhan, and on the other hand, Tesla continues to default on its plans, postponing the planned RoboTaxi launch conference from August to October, and its market value plummeted by 500 billion yuan overnight.
Do you still remember the self-driving technology that Baidu CEO Robin Li first demonstrated on the Fifth Ring Road? The self-driving technology that was investigated by the traffic police at that time has now become a formal commercial project and is no longer considered a violation of traffic regulations.
This makes people curious about what technology is behind it. Today, let’s take a look at Turnip Run and Baidu Apollo behind it.
What is the relationship between LuoBoKuaiPao and Apollo Intelligent Driving?
Apollo Intelligent Driving and Carrot Run are two key projects of Baidu in the field of autonomous driving. They are closely related, but have different focuses.
Apollo Intelligent Driving
Apollo Intelligent Driving is Baidu's autonomous driving technology platform, covering the research and development, testing and deployment of autonomous driving vehicles. It provides a complete set of open source autonomous driving solutions, supporting various autonomous driving functional modules including perception, decision-making, planning and control. The Apollo platform aims to promote the development of autonomous driving technology and build an open autonomous driving ecosystem with automakers, suppliers and other partners.
Carrot Run
Apollo Go is an autonomous driving travel service launched by Baidu based on the Apollo intelligent driving platform. It is mainly aimed at ordinary consumers and provides unmanned taxi (Robotaxi) services. This service has been launched in many cities, including Beijing, Shanghai, Guangzhou, Chongqing, etc. Users can book and use unmanned taxis through mobile applications.
"Robot Run" has launched a 24-hour service in Wuhan. Baidu's driverless fleet has expanded to 300 fully driverless vehicles, and the operating area has expanded to 3,000 square kilometers, covering half of Wuhan's population. This large-scale deployment and wide coverage have significantly improved the availability and convenience of the service.
Through continuous technological innovation, Apollo Go has improved the safety and reliability of driverless technology. By April 2024, Apollo Go had safely traveled more than 100 million kilometers without any major accidents (Enterprise Technology News and Analysis). This not only demonstrates the maturity of the technology, but also enhances the public's trust in driverless services.
Relationship and difference
Technical foundation:
The technical foundation of LuoBoKuaiPao is derived from the Apollo intelligent driving platform. Apollo provides the core technologies and solutions required for autonomous driving, while LuoBoKuaiPao applies these technologies to actual travel services.
Application scenarios:
Apollo Intelligent Driving mainly focuses on technology research and development and ecosystem construction, serving R&D personnel and enterprises. Luobo Kuaipao is oriented to end consumers, providing actual driverless travel services.
Objective:
Apollo Intelligent Driving aims to promote the development and popularization of autonomous driving technology and establish an open cooperation platform. The goal of Carrot Run is to provide safe, efficient and convenient driverless travel services and enhance users' travel experience.
In summary, Apollo Intelligent Driving is the core platform of Baidu's autonomous driving technology, and LuoBoKuaishou is a practical application service based on this platform. They complement each other and jointly promote Baidu's development in the field of autonomous driving.
Apollo Intelligent Driving Development History
Baidu Apollo has three business models: providing autonomous driving solutions to OEMs, Baidu manufacturing cars, and driverless taxis.
(1) Providing autonomous driving solutions for OEMs:
Based on its long-term AI technology accumulation and big data analysis, Baidu provides 9 complete solutions for OEMs: shared unmanned vehicle solutions, Minbus autonomous driving solutions, Valet Parking autonomous driving solutions, intelligent signal control intelligent transportation solutions, vehicle-road collaborative intelligent transportation solutions, Xiaodu in-vehicle OS Internet of Vehicles solutions, Internet of Vehicles intelligent car cloud solutions, 5G cloud driving service solutions, and Apollo Air plan. It empowers multiple OEMs and builds the world's largest autonomous driving ecosystem.
(2) Baidu's car manufacturing:
Baidu and Geely jointly established Jidu Auto, with Baidu holding a 55% stake and 3/5 of the board seats. It is expected to invest about RMB 50 billion in the next five years, with a core team of 200 people, which is expected to be expanded to 3,000 people by 2022. Jidu Auto plans to combine Apollo's intelligent driving technology with the electric car manufacturing SEA intelligent evolution experience framework to develop the L4 autonomous driving framework JET, and is expected to launch a new car to the market every year based on this framework.
(3) Driverless taxis:
Based on the Apollo platform, Baidu has a first-mover advantage through core technologies in the four areas of vehicle, road, cloud, and map. It is currently the Robotaxi company with the earliest layout, the most patents, and the strongest financial resources in China. It is expected to open its business to 30 cities in 2023, deploy 3,000 vehicles, acquire 3 million users, and realize a recyclable business model in the region.
Apollo platform development history:
Apollo 1.0 (April 2017): Provides infrastructure and functional modules, mainly used in closed scenarios.
Apollo 2.0 (December 2017): Supports autonomous driving capabilities on urban roads and highways.
Apollo 3.0 (July 2018): Achieved vehicle-road collaboration and started commercial pilot operations.
Apollo 3.5 (December 2018): Enhances autonomous driving capabilities in complex urban environments.
Apollo 4.0 (July 2019): covers all scenarios and adopts multi-level safety redundancy design.
Apollo 5.0 (June 2020): Large-scale deployment and testing.
Apollo 6.0 (December 2021): Commercial operation of driverless taxis.
Apollo 7.0 (July 2022): Introduces deep learning models and the Apollo Studio online development platform.
Apollo 8.0 (January 2023): Provides an extensible software framework and upgraded simulation services.
Apollo 9.0 (December 2023): Optimize the software framework interface, upgrade the detection model, and introduce an incremental training method.
As a new, professional, and local Tier 1, Baidu's Intelligent Automobile Division has formed a competitive, complete automotive intelligent product matrix covering "driving, cabin, and mapping", which can meet the various needs of domestic and foreign automakers for automotive intelligence and help automakers build future-oriented smart cars.
-
In terms of intelligent driving, Baidu has launched a full range of intelligent driving products covering low, medium and high combinations, including Apollo Parking, Apollo Highway Driving Pro, and Apollo City Driving Max. These products have been mass-produced on many star models.
-
In terms of smart cabin, Baidu launched the industry-leading smart cockpit solution, including Xiaodu in-car OS and Xiaodu in-car voice SDK;
-
In terms of cockpit integration, Baidu is also working with car companies to create cockpit integration solutions;
-
In terms of intelligent maps, it provides automobile companies with products such as in-vehicle navigation maps, high-precision maps, and human-machine co-driving maps.
At present, Baidu's automotive intelligent solutions have been widely recognized and have been mass-produced on 134 models of 31 automobile brands, with a total of over 7 million vehicles installed.
The specific logic levels of Baidu Apollo are as follows:
-
Functional layer: Through the customization of interactive logic (including HMI and state machine, etc.), the optimized customized experience of related functions can be achieved. In addition, car companies can also develop cross-domain functions on their own based on the algorithm atomization capabilities provided by Baidu, such as the magic carpet function, to flexibly meet the needs of different car companies.
-
Decision-making planning: In order to meet the car companies' ultimate pursuit of user experience, Baidu provides up to 12 major categories and nearly 50 product calibration parameters, allowing car companies to customize the effects of intelligent driving products in a convenient way, thereby better configuring driving styles.
-
Vehicle control: Automakers can choose to integrate their own vehicle control modules into the solution according to interface requirements based on their own circumstances.
-
Bottom software layer: Provides a unified computing acceleration service interface to decouple intelligent driving algorithms and chip platform selection.
-
Hardware layer: The solution supports a variety of sensor combinations, and car companies can independently balance the cost-effectiveness and choose the hardware solution that best suits them.
-
Data closed loop: It can support car companies to configure personalized data collection and algorithm strategies.
Hardware development platform support includes computing, sensors, remote cloud, drivers and other partner support.
The real fire is to run away quickly
Next, let’s talk about Carrot Run.
As an important scenario for the commercialization of autonomous driving, Robotaxi is an autonomous driving ecosystem based on efficient and safe travel services. Its commercialization requires the ability to operate large-scale travel services. In addition to factors such as traffic, operation, experience, scheduling, and service, the successful operation of Robotaxi services also requires the improvement of autonomous driving-related policies, which is a prerequisite for the large-scale commercialization of autonomous driving vehicles. The cultivation of user habits and the number of market participants affect the formation of economies of scale. The cost of autonomous driving determines the market competitiveness of Robotaxi products, and the maturity of technology determines whether autonomous driving vehicles are safe and convenient and consumers' acceptance of autonomous driving services.
Take stock of Baidu's autonomous driving models over the years:
Baidu's self-driving cars are safer than human drivers after 10 years of hard work
In May 2024, Baidu Apollo released Apollo ADFM (Autonomous Driving Foundation Model), the world's first large-scale model that supports L4 autonomous driving. At the same time, it also launched the sixth-generation unmanned vehicle, the Carrot Run, which is equipped with Baidu's sixth-generation intelligent system solution and priced at only 200,000 yuan.
Baidu Apollo ADFM reconstructs autonomous driving based on large model technology, which can take into account the safety and generalization of the technology, making it more than 10 times safer than human drivers, and achieving full coverage of complex scenes in the city. It is precisely relying on the application practice of large models of autonomous driving️ that Baidu Carrot Run has conquered the complex road scenes in Wuhan, achieving full coverage of scenes in the entire city of Wuhan and all time and space, providing convenient unmanned travel services for nearly half of Wuhan citizens. At the same time, in the field of L2+ intelligent driving, ANP3, the only pure visual city pilot assisted driving product in China, will also fully apply the large model of autonomous driving Apollo ADFM, and will be upgraded to ASD (Apollo Self-Driving). It will soon be mass-produced and launched in the entire series of Jiyue models, taking the lead in realizing intelligent driving that can be driven across the country and anywhere there is a Baidu map.
Xu Baoqiang, general manager of Baidu's autonomous driving technology department, said: "Based on the large-scale autonomous driving model, it only takes half a year to complete the preparation of fully autonomous driving travel services in a new city. In the future, this verification time will be further shortened."
The Carrot Express unmanned vehicle, equipped with Baidu Apollo's sixth-generation autonomous driving system solution, fully applies the "Baidu Apollo ADFM large model + hardware products + security architecture" solution. This solution ensures the stability and reliability of the vehicle through a 10-fold safety redundancy solution and a 6-fold MRC safety strategy, and its safety level is close to that of the domestically produced large aircraft C919.
The sixth-generation unmanned vehicle of Carrot Fast Run is equipped with a dual-computing unit with high computing power at the automotive level, with a computing power of up to 1200 Tops. The 38 sensors of the whole vehicle are deeply integrated to achieve full coverage of the three-fold detection capabilities of long, medium and short distances, with more accurate perception capabilities. At the same time, it has seven-fold full redundancy systems including architecture redundancy, computing unit redundancy, and braking system redundancy. If any single component or system fails, the spare redundant system can instantly complete the replacement; in terms of software, it is equipped with a fault diagnosis and risk reduction system integrating the whole vehicle + intelligent driving system.
At the press conference in May, Baidu also introduced end-to-end autonomous driving technology, which may also be used in Carrot Run. Its end-to-end architecture is a two-stage end-to-end architecture, with two-stage networks for perception and decision planning, and end-to-end autonomous driving is achieved through invisible transmission and joint training.
On the perception side, visual image data is obtained through the camera, and active light sensor data is obtained through the lidar and 4D millimeter wave radar.
Then the perception network is used to identify visual and active light features, and the BEV features are integrated to determine road features, obstacles, etc.
Finally, end-to-end autonomous driving is achieved through joint training through explicit three-dimensional vector space (using a three-dimensional coordinate system to accurately represent the position and direction of an object) and implicit BEV features (from the BEV perspective, all information in the three-dimensional space is not directly given, but through some indirect methods, such as feature encoding or deep learning models to implicitly represent and infer this information).
As of April 2024, Baidu Apollo's autonomous driving mileage has exceeded 100 million kilometers, but there has never been a major accident with casualties. However, data from the past two years show that the actual vehicle accident rate is only 1/14 of that of human drivers.
Market exploration
Users and markets are the only criteria for testing technology. In 2021, Baidu Carrot Run began its market exploration and established the world's largest unmanned vehicle automatic operation network in practice, which can achieve global monitoring of the entire process, including automatic maintenance, risk warning, automatic OTA, remote control windows and more than 30 other fully automatic management units.
The sixth-generation unmanned vehicle of LoBo Express, equipped with Baidu Apollo's sixth-generation intelligent system solution, has a 60% reduction in vehicle cost compared to the fifth-generation vehicle. The price is only 200,000 yuan, which has once again set a new industry record.
Wang Yunpeng said that Baidu and its car company partners will continue to think about and define driverless cars, with the goal of reducing costs by half and increasing capabilities tenfold with each generation of products.
Final Thoughts
Baidu has made significant progress in the field of autonomous driving technology through Carrot Run and Apollo Intelligent Driving. It has also enabled L4 autonomous driving to be more widely tested and recognized by the market. At least for now, people's demand for technology is very strong.
· END ·
You are welcome to set us as “
Star
” so that you can receive push messages as soon as possible.
Scan the QR code to follow: Automotive Development Circle, reply "
Auto
"
Get free information packages on Autosar architecture, standards and systems, application development, etc.!
Scan the QR code to add the assistant and reply "join the group"
Exchange experiences with automotive developers face to face
京公网安备 11010802033920号