Apollo 5.0 Is Coming. Baidu Has Become an "old Driver"!
Last week, baidu quietly released the latest version of Apollo, its open source full stack software solution for driverless cars. Following Apollo Lite, Apollo 5.0 introduced a series of new features, including upgraded perception models for "complex" road scenes and customized sensor calibration services.
"Apollo 5.0 is designed to support the mass production of geofence autopilot," Baidu pointed out in the attached GitHub change log. "This car now has a 360 degree view... Making it safer and more conscious."
Apollo 5.0 also introduced a new data pipeline service, including calibration options for each vehicle, as well as a neat prediction evaluator and map data verification tool. Highlights are the open space Planner - a new scenario based planning algorithm - and better support for parking and intersections, including distinguishing between parking signs and traffic lights. Finally, dreamland, Apollo's web-based simulation platform for model validation and testing, has been updated with more powerful scene editor and control loop simulation.
Development of Baidu Apollo program
Since the launch of Apollo program, it has gone through a long road of about two years. At the 2018 consumer electronics show, baidu released the second major iteration of Apollo - Apollo 2.0, which introduced a new reference vehicle, an encrypted air update framework, an improved computer vision algorithm, and a system that can better determine the driving position of the vehicle.
In April 2018, baidu finally unveiled Apollo 2.5, which realized vision based perception improvement, real-time relative mapping, new driving scenes and visual debugging tools. At the same time, Baidu has launched an automobile safety laboratory in cooperation with China Automotive Technology Research Center and China information and Communication Research Institute.
Apollo 3.0 adds support for valet parking, autonomous mini bus and autonomous mini car, as well as integration with Baidu voice activated telematics software, which can perform facial recognition and monitor signs of driver fatigue. In addition, the responsibility sensitive security (RSS) of Mobil international, a subsidiary of Intel, was also included in Apollo's code base. The so-called responsibility sensitive safety is a "common sense" road decision-making method, which incorporates good driving habits (such as maintaining a safe following distance and giving other vehicles the right of way).
Apollo 3.5 introduced the ability to complete unprotected turns (as we all know, this is a challenging action of driverless vehicles); Manage speed bumps; And unobstructed areas, side passages, narrow lanes and parking lots. In addition, it marks the launch of Apollo enterprise (a set of autonomous and interconnected services for mass-produced vehicles) and Apollo network real-time framework (a high-performance runtime compatible with lidar sensors such as velodyne's vls-128).
Baidu's agility helps it keep pace with NVIDIA. Earlier this year, NVIDIA launched a new full stack automatic driving solution on drive autopilot. A newer challenger is Japan's tier IV, which recently raised $28 million to develop an open source software platform for driverless cars, autoware.
The Apollo program advanced by leaps and bounds
At present, Apollo's code has increased to 400000 lines, more than twice the 165000 lines announced by Baidu in January 2018. Now it is being tested, contributed or deployed by Intel, NVIDIA, NXP and more than 130 global partners. (up from 116 partners in July 2018.) according to Baidu, the number of developers obtaining Apollo code from the GitHub repository of the project is 12000, an increase of 20% over the middle of 2018.
With the continuous progress of technology, baidu also has more and more partners. Udelv, headquartered in California, said in January that it would deploy up to 100 Apollo 3.5 automatic vehicles to American cities including the San Francisco Bay Area in 2019. The other Apollo plan adopters include Volvo and Ford, which both promised to test Apollo driving autopilot on China's roads in 2019.
Baidu also cooperated with Chinese carmakers Chery, BYD, great wall, Hyundai Kia, Ford and vmmotori to launch Apollo enterprise automobile solutions. FAW Group, which develops Hongqi series luxury cars, is also another close partner. Last year, the group announced plans to launch a "limited number" of Apollo cars in China next year. According to the foreign media Japan economic news, Toyota will also join Baidu's automatic driving development alliance "Apollo program" in the near future.
According to Baidu's official website data and public reports, including Toyota, nearly 150 automobile related enterprises have joined the Apollo program. The platform will disclose the large-scale automatic driving data set Apollo scape, as well as the latest software and hardware technical information in vehicle networking and automatic driving, and each enterprise will promote the specific development respectively. At the same time, it also needs to extract the desensitization data of participating manufacturers in the R & D process, so as to optimize and iterate its own technology.
In addition, Baidu's automatic driving project has also made progress. On July 1, the Beijing automatic driving test management joint group released the first batch of T4 level automatic driving test licenses, a total of 5, all of which were in Baidu's pocket. T4 license is the open road test qualification certification with the highest technical level, the highest standard and the most difficult test scenario in China. The test and evaluation of autonomous vehicles is divided into T1 to T5, of which T1 is the most basic straight road, with only simple traffic settings such as traffic lights; T2 is a simple urban scene, which allows autonomous vehicles to turn right; T3 is a common urban scene with urban plane overpass; T4 is a complex urban scene, with tunnels, boulevards and other settings; T5 is a special urban scene, which can realize the complex traffic and weather environment such as rain and fog, wet road surface and so on. In addition, pony.ai, Tencent and didi have obtained T3 level licenses, Daimler, Audi and Weilai are T2 level, and BAIC new energy is T1 level.
Baidu plans to achieve "complete autonomy" on expressways and urban roads by 2020. Its competitor pony.ai in Beijing has also raised US $214 million in venture capital and launched a pilot driverless taxi in Guangzhou in early April. Meanwhile, alphabet's waymo said that it currently serves more than 1000 drivers and has more than 600 vehicles. GM's cruise automation has been testing automated taxi services for employees in San Francisco and plans to launch public services this year.
