Baidu Apollo 9.0 introduces autonomous driving open platform with 120,000 lines of code refactored and adapted to ARM architecture for the first time

Gamingdeputy reported on December 19 that Baidu officially launched a new upgraded version of the Apollo open platform today—— Apollo Open Platform 9.0.

Zhang Liang, general manager of Baidu’s autonomous driving platform ecosystem department, summarized the main highlights of this upgrade: “Apollo open platform 9.0 has achieved comprehensive upgrades in terms of engineering, algorithms and tools., the general layer can enable large-scale implementation of various application scenarios, the overall operation is more flexible and easy to use, and the use scenarios are universal and easy to expand. While greatly improving development efficiency, it can help more developers quickly build their own autonomous driving systems. “

Advertisement

In terms of engineering framework, in order to enable developers to assemble autonomous driving applications more flexibly and facilitate secondary development, Apollo Open Platform 9.0 comprehensively upgrades package management, splitting modules into smaller software packages according to the granularity of functions. Developers can more conveniently choose to use it according to their own needs. It also provides a wealth of functional components and plug-ins, and has improved and optimized functional extensions.

Based on this, after the unified scheduling interface, developers can complete the scene demo construction in as soon as 1 day. The simplified parameter adjustment method doubles the parameter adjustment efficiency. The new plug-in mechanism reduces the code learning cost by 90% and the code volume by 50%. %, greatly improving Apollo's secondary development capabilities.Apollo open platform 9.0 also adapts to the ARM architecture for the first time.

In terms of algorithms, Apollo Open Platform 9.0 optimizes the perception algorithm based on the previous one.Lidar detection uses the relatively new CenterPoint model, and visually uses the Yolo X + Yolo 3D model.pouring millions of data into these two models for training, the recall rate and precision rate are said to have “significantly improved.”

Apollo open platform 9.0 also provides incremental training,Support independent completion of model training, which can improve the detection capabilities of special targets and special scenes while maintaining the original detection capabilities of the model, thereby easily improving the detection effect of customized scenes at a lower cost.In addition, the new platform fully supports 4D millimeter wave radarwhich enhances obstacle detection and extreme weather scene safety.

Advertisement

picture

In terms of tools, in addition to adding new functions based on the original tools,Apollo open platform 9.0 also adds new tools including high-precision mapping, sensor calibration and integration. The newly upgraded Dreamview+ of Apollo Open Platform 9.0 has been comprehensively improved in multiple aspects such as multi-scenario use, free layout, and data resources. The debugging process is simpler, the window layout is more flexible, and resource access is more convenient.

In addition, Apollo Open Platform 9.0 also reconstructs the document platform to achieve more convenient operation, smoother reading, and more substantial content, effectively reducing developers' learning and usage costs.

At this press conference, Dr. Dong Zhaozhi, Executive Vice President of Kaiwo Group and President of Academia Sinica, Qiang Xiaowen, Secretary of the Party Committee, Deputy Headquarters and Director of the Product Research and Development Institute of Dongfeng Motor Co., Ltd.’s Commercial Research Headquarters, King Long United Automotive Industry (Suzhou) ) Co., Ltd.'s Forward-Looking Technology Research Institute, Dr. Liu Mingchun, and Dr. Liu Junchuan, Vice President of TZN Technology Co., Ltd., were also present.

Baidu said that Apollo Open Platform 9.0 has powerful scene-wide capabilities and is easier to use and expand;The adaptation process is reduced by 40%, the amount of code reading is reduced by 90%, and the amount of code debugging is reduced by 80%., it can be used “out of the box” and the closed-loop self-driving vehicle can be completed within a week. At the same time, Apollo Open Platform 9.0 covers more scenarios and can complete the closed-loop operation of the scenario application business system in one month. On this basis, the sensor calibration and map creation cycles have been shortened to hours, significantly shortening the implementation time and making it more efficient.

picture

at the same time,Apollo open platform 9.0 has lower hardware costs and more optionshardware selection supports 3Lidar+4Camera, the camera supports more than 4 manufacturers, upgraded from USB3.0 to GMSL, lidar has added 32-line, 64-line and other multi-brand and multi-model equipment, and positioning equipment has added more than 3 mainstream brand equipment Support can fully assist the implementation of large-scale applications in different scenarios.

picture

Gamingdeputy learned from the press conference that currently, the Apollo open platform has brought together more than 160,000 developers from more than 170 countries and regions around the world.During the development process of Apollo open platform 8.0 to 9.0, 120,000 lines of code were refactored and 200,000 lines of code were added..

Advertisement