Our reporter Yang Qingqing reports from Beijing
On July 4th, at the 2018 Baidu AI Developer Conference, the 100th autopilot bus was broadcast live on the big screen. “Apolon” was in the process of the Xiamen production workshop, “Apolon” was also regarded as a global The first L4-class production of unmanned vehicles.
Despite the fact that around this production car, the industry is still controversial: Apollo is a commercial vehicle, not the passenger car that is usually expected; the difficulty of driving in a closed park and on a fixed line is completely different from the open road …&hellip But in any case, this is still the highlight moment of a Baidu unmanned vehicle. Otherwise, the scene will not applaud the thunder, the news will not be screened, "Apollon" manufacturer Jinlong Automobile will not stop the price in the near-line seal in the afternoon.
Supporting this “Apollon” production car is the updated version of Apollo 3.0. According to Li Zhenyu, vice president of Baidu and the general manager of the intelligent driving business group, “Apolon” is equipped with “the solution of the surface vector production”, which has become the core of Apollo 3.0. Based on this solution, developers on the Apollo platform can transform their own unmanned vehicles in as little as three months.
In addition to the solution, there are many upgrades to the Apollo 3.0 technology framework. Previous hardware reference platforms were upgraded to hardware development platforms, sensor units were released to support a larger number and type of interfaces, and vehicle reference platforms were upgraded to open vehicle certification platforms and open vehicle interface standards.
“The key word for Apollo 3.0 is face vector production, which is more open. Li Zhenyu said that Apollo has opened more than 220,000 lines of code, and more than 10,000 developers recommend Apollo's open code. The number of eco partners is 116, “Let every developer get intelligence equally and conveniently.” The ability to drive is the philosophy that Apollo has always insisted on. ”
Technical framework upgrade
Since the release of Apollo 1.0, this is the fifth iteration of Apollo.
According to Wang Jingao, head of Apollo platform development, Apollo 3.0 has designed a new platform architecture compared to previous generations. In the Apollo 1.0 version architecture, the middle layer is the open source software platform and the upper layer is the cloud service platform. Below these two levels are the hardware reference platform and the vehicle reference platform. Since the release of the version, the platform capability module has been gradually illuminated and has been fully lit.
“In today's Apollo 3.0 version, the vehicle reference platform is upgraded to a vehicle certification platform, the hardware reference platform is upgraded to a hardware development platform, and the cloud service platform is added with a mass production service suite. Based on this, a Valet Parking is provided. MicroCars, MiniBus, and Small Car OS four mass production solutions. ” Wang Jingao said.
On the upgraded hardware development platform, Apollo first released the Appollo Sensor Unit (ASU), including laser radar, car camera, millimeter wave radar, integrated navigation system, ultrasonic equipment, etc. Plug and play on the Apollo platform.
At the same time, the hardware development platform also adds the underlying software abstraction layer, which can adapt to multiple data formats, define common API interfaces, realize sensor time synchronization, and provide real-time monitoring of hardware devices to seamlessly connect upper-layer software and devices. drive.
As for the upgraded vehicle certification platform, a total of 17 indicators including functions, performance, safety, and energy consumption are opened from the two aspects of the line control system and the vehicle system. Also open tools and documentation including DBC file conversion tools, dynamic curve calibration tools, instruction security testing tools, interface requirements documentation.
In Li Zhenyu's view, Apollo upgraded the vehicle reference platform to the vehicle certification platform, based on Baidu's past experience, linking the needs of car companies and developers. “In the past few years, Baidu has experimented with many models and has stepped on a lot of pits in the process. “Lin Zhenyu said, “We have summarized the past pits into experience, summed up the 17 categories of indicators, and opened them as an Apollo vehicle verification platform. ”
This also means that the vehicle certification platform can link the needs of car companies and developers, thus accelerating the deployment of driverless. “ Depots can refer to these indicators to develop their own models. For developers, you can refer to these indicators to choose the right autonomous vehicles. More importantly, by understanding these indicators, you can know which car is strong for autonomous driving. & rdquo; Li Zhenyu explained.
Unmanned car mass production era
In addition to upgrading the two platforms, and more open, "face vector production" is another key word in this Apollo 3.0 version. In Wang Jingao's view, Apollo 3.0 realized the first open plan from technology research and development to mass production, and entered the era of automatic driving mass production.
It is worth noting that this time Apollo launched a mass production program for autonomous driving for low-speed campus scenarios. Li Zhenyu said that in the mass production process for autonomous driving, Apollo 3.0 advocates starting from the low-speed park and promoting the steady landing of autonomous driving in a gradual way, “because safety is the most important. ”
Bai Yongyi, general manager of Baidu's Autopilot Division, said that “Automatic Shuttle Bus MiniBus” and “Unmanned Cars MicroCar” are equipped with sanitation vehicles, buses, vans, vans, and mini-cleaners. , micro-logistics vehicles and other vehicles. In the future, Apollo will cooperate with Beijing Environmental Sanitation Group to build a medium-sized and medium-sized sweeping vehicle of mass production grade. It is expected to achieve a full industrial chain from production to operation in 2019; Apollo will also cooperate with Suning Logistics to achieve a logistics automatic driving operation, which will be realized in 2020. Urban terminal logistics automatic driving distribution.
Apollo platform upgraded seven capabilities at the same time, including low-speed campus sensing algorithms, low-speed park planning algorithms, low-speed park control solutions, energy production safety monitoring, and energy-generating HMI debugging tools. Energy production developer interface and developer contribution relative map. Around the four key components of the autopilot kit, mass production tool components, safety assurance system, and operational scheduling solutions, the Apollo platform also has a corresponding solution.
For the Internet of Vehicles, Apollo has introduced a small-scale in-vehicle OS production solution. The data shows that the small-scale in-vehicle OS consists of four parts: the LCD instrument panel component, the streaming media rearview mirror assembly, the large-screen smart car component, and the small-scale in-vehicle robot component. It is possible to transform a traditional car into a smart car in a day.
However, these mass production solutions that currently focus on the park's automatic driving ability are still just starting. “In 2019, Apollo will unlock auto-driving and mass production limited areas for automatic driving in restricted areas. By 2020, mass production will be launched in simple urban roads. In 2021, high-speed and urban roads will be automatically driven. & rdquo; Wang Jingao said, & ldquo; This is still a process that needs continuous expectation. ”