Baidu's "complete open automatic driving ecosystem" in a closer examination
SHENZHEN, China — The concept of self-driving cars is not new. As early as 1939, GM floated the idea of cars driving themselves in its Futurama exhibition at the World’s Fair in New York.
It wasn’t until the 1990s or much later, though, when self-driving cars started to become more realistic thanks to breakthroughs in computer processing power, artificial intelligence, sensors, and robotic controls. Google, Intel, Uber, Tesla, Ford, GM, and BMW have all been scrambling to take a lead in the nascent field of autonomous vehicles (AVs).
Many industry observers have long assumed that AV is a domain that foreign-owned companies will dominate. In fact, Chinese companies didn’t even start to come into the field until a few years ago. But to the surprise of many players abroad, Apollo, an autonomous vehicle platform that Chinese internet giant Baidu launched only a year ago, quickly gained recognition and has grown to be the leading player in the AV industry.
Apollo: ‘Android for automated vehicles’
On April 19, 2017, Baidu officially announced the Apollo plan, billed as the first system-wide opening of the global automated driving technology. Baidu says that Apollo is “a complete open automatic driving ecosystem” that can help partners in the automotive industry and autonomous driving to combine vehicle software and hardware systems to quickly build their own complete AV system.
The Apollo platform consists of three parts: localization, open-software platform, and cloud-service platform. Baidu explains on its Apollo website that the Apollo platform provides partners with high-precision map services with advanced technology, extensive coverage, and high automation. Apollo also offers a simulation engine, which the company claims as “the only one in the world that is open and is equipped with massive data.” Moreover, Apollo’s end-to-end autonomous driving algorithm has “the world’s largest volume of deep-learning data sets” that are open.
For Apollo to become a complete AV platform vendor, it also needs chips, sensors, vehicle architecture, and other pieces of hardware to support Apollo’s software algorithms. Therefore, Baidu chose to cooperate with auto suppliers and OEMs both at home and abroad. As Baidu provides its software platform, auto suppliers offer hardware integration, productization, and manufacturing facilities where final products can be produced.
By leveraging these hardware modules produced by its suppliers, Baidu offers a “reference vehicle platform” including computing units; "reference hardware platform" consisting of sensors such as GPS, cameras, and lidar; and human-to-machine interface devices. While Baidu won’t be directly involved in the production of hardware, it will offer services including the reference hardware platform and reference vehicle platform. The mass production of AVs based on the Apollo platform will be left to auto manufacturers.
With the support of the Apollo platform, partners can develop, test, and deploy autonomous vehicles faster. As more partners get involved, Apollo expects that they will be able to accumulate and amass more driving data. Compared with a closed-system approach opted by foreign manufacturers, Baidu believes that Apollo can mature AV technology at a faster rate, giving each participant more benefits.
Apollo: ‘Android for automated vehicles’
Baidu intends to build Apollo as “Android for automated vehicles.” By using the same “open-source” playbook that helped Android to become the most dominant operating system platform in the global smartphone market, Baidu is betting on Apollo, hoping that it will take the global AV industry by storm. Tech communities in Europe and the United States are reportedly already applauding Apollo’s pioneering efforts to push the open-source practice into the AV industry.
Milestones of the Apollo plan
Baidu has been investing heavily in the research and development of self-driving cars since 2015 — well before it unveiled its Apollo plan. In December 2015, Baidu conducted full-AV driving tests on both highways and urban roads in Beijing.
In September 2016, Baidu obtained a road test license for driving autonomous vehicles in California, and in November of the same year, Baidu launched an open-circuit driverless car trial operation in Wuzhen, Zhejiang.
On April 19, 2017, Baidu officially announced the Apollo platform. At that time, the industry commented highly, but Baidu did not release anything substantial.
On July 5, 2017, Baidu released Apollo 1.0, an important milestone in the Apollo project. In the 1.0 release, Baidu made available the company’s very valuable data including closed-track automated driving and end-to-end self-positioning.
On Sept. 20, 2017, Baidu released Apollo version 1.5. Its focus was on opening five core competencies, including object awareness, decision-planning, cloud simulation, high-precision map services, and end-to-end learning. Among them, the first four capabilities were made openly available for the first time. The self-driving car built on the 1.5 version can support both day and night fixed-lane auto-driving, enabling AVs to identify objects in the night environment and other obstacles in atypical traffic scenarios. Some media claimed that this was what Apollo 1.0 should have had.
On Jan. 9, 2018, Baidu launched Apollo version 2.0, which supports automated driving on simple urban roads. Apollo 2.0 also introduced a “scenario-based,” commercial driverless solution. Four major AV processing platform vendors including Intel, Nvidia, NXP, and Renesas now support Apollo 2.0. The Apollo platform now provides four modules including cloud services, software platforms, reference hardware platforms, and reference vehicle platforms.
On April 19, 2018 — the first anniversary of the announcement of the Apollo plan — Baidu released Apollo version 2.5. In the latest version, Baidu offers four major updates:
1) It provides more driving scenarios. Apollo 2.5 supports limited-area visual high-speed automated driving, unlocking freeway scenes; open vision perception; real-time relative maps; high-speed planning; and control capabilities.
2) It offers a lower-cost sensor solution. With a camera-based vision-aware solution, the cost of the sensor can be reduced by 90% compared to the previous solutions, significantly lowering the threshold for autonomous driving research.
3) It added truck logistics application scenarios. It can now support passenger cars, trucks, buses, logistics vehicles, sweeping vehicles, and other models.
4) It offers new development tools including Dockerfile, DreamView visualization tool, Apollo Drive Event data collector, Apollo high-precision map data collector, and Apollo Cloud Simulator. These newly released tools will help improve developers’ R&D efficiency, according to Baidu.
According to the plan, Apollo will be updated to version 3.0 in 2018 and version 4.0 by 2019.
Apollo has 100 Partners
At the Apollo anniversary event, Li Zhenyu, vice president of Baidu and general manager of the smart driving business group, announced the 100th partner of the Apollo project: BYD.
The industry observers weren’t surprised at the cooperation between the two companies. That’s because during the New Year’s celebration this year, China Central Television (CCTV, China’s state television broadcaster) staged a self-driving car show on the Hong Kong–Zhuhai–Macao Bridge, featuring Baidu and its automatic driving partner with BYD thrown into the mix.
Apollo’s official website shows a long list of its partners, which includes many big shots from tech and automotive industries based in China and around the world: Ford, Daimler, BAIC, FAW, Chery, BYD, Jinlong Bus, Bosch, Continental AG, and ZF; sensor and chipmakers such as Velodyne, Intel, and Nvidia; server suppliers such as Microsoft; transportation service provider Shouqi Limousine & Chauffeur; autonomous driving system suppliers idriverplus and Momenta; and others.
Eighty percent of Apollo’s partners are autopilot parts, chips, radar, camera manufacturers, and research institutes, while the remaining 20% are automakers. These partners are endorsing Apollo’s ecosystem. For example, Baidu and Bosch, the global giant in ADAS, signed last year a “strategic cooperation framework agreement on smart mobility in China.” The two companies signed the deal under the witness of China’s Premier Li Keqiang and German Chancellor Angela Merkel.
Apollo’s other key partners include Nvidia, the leader of artificial intelligence chips, and Velodyne, the originator of lidars.
Apollo Ecosystem has expanded its scope of cooperation with OEMs, tier ones, core suppliers, travel service providers, startups, investment funds, relevant governments, and research institutions. Thirty percent of them are companies and institutions outside of China.
Why is Baidu’s Apollo attracting so many heavyweights in the global industry?
For one, China represents one of the largest markets for AVs. Baidu is also an exceptional partner in China. The company has significant advantages in three key technologies essential in AVs: technologies tied to automotive industry, artificial intelligence, and mapping.
Considering that the Chinese automotive industry has trillion-dollar potential in the era of autonomous driving, there is no company better than Baidu to reap the fruits.
Baidu’s level of artificial intelligence is second only to AI efforts by leading tech companies such as Google, Facebook, and OpenAI, a non-profit artificial intelligence research company based in San Francisco.
It’s hard for those global rivals to compete against Baidu in China.
Baidu is one of 13 companies armed with high-definition map licenses. As previously reported by foreign media, any company with a plan to launch self-driving cars in China must cooperate with a licensed map company.
Revolution in the field of automated driving
Automated driving is a large and complex project involving many technologies. The core principle of AV technology is to capture images of road conditions, make judgments, and then control the movement of the vehicle. Hence, the AV is an integrated system featuring environmental awareness, planning and decision-making, and multi-level assisted driving functions. It has extremely high technical requirements for artificial intelligence and automated control.
Judging from different players in the field of automated driving, traditional car OEMs are good at building cars, but they lack experience in a range of technologies and services such as software, algorithms, and internet content. Tech companies are just the opposite. For example, companies such as Audi and BMW aren’t developing AV technologies by themselves but in close collaborations with such vendors as Delphi and Mobileye. Companies such as Google’s and Baidu’s self-driving cars are reusing the chassis of traditional cars and retrofitting them with new technologies.
Automotive is a century-old industry. It is difficult for internet companies to build cars on their own. The best approach is to cooperate with auto manufacturers and gradually promote the future of automated driving and intelligence.
Baidu’s plan for Apollo is in line with this approach. Apollo brings in companies that have mastered the core technologies of autonomous driving. Through collaboration, Apollo has developed a complete set of automated driving solutions.
Let the future vehicle manufacturers develop their own AV systems based on Apollo, just as Android allowed companies who have never been in the mobile phone business to grow to be leading players in the smartphone market.
Apollo’s AV platform allows automakers to quickly build a complete autonomous vehicle system of their own. For the entire industry, this will reduce the R&D costs of self-driving cars, allowing more car manufacturers to participate and promote the rapid adoption of autonomous driving technology.
Furthermore, at a time when the AV development field is still in chaos, automated driving solutions based on Apollo will be of great significance because they can offer opportunities to unite technical standards. Many protocols and interfaces can be unified under the standard of a single-package solution. This will not only lay the foundation for future smart transportation and the networking of smart cities but may also become the key to the popularity of self-driving cars.