Beijing (Gasgoo)- On December 19, Baidu launched the upgraded Apollo Open Source Platform 9.0, a significant advancement aimed at developers and ecosystem partners. This new version fortifies its leading edge in autonomous driving development with enhanced algorithm capabilities, a more user-friendly toolkit, and an easily expandable range of common scenarios.
Since its inception, the Apollo Open Source Platform has maintained rapid iterations, continually breaking through boundaries—from engineering framework optimization to developer accessibility and implementation of universal scenarios.
In terms of the engineering framework, the Apollo Open Source Platform 9.0 update meticulously restructures package management, fragmenting modules into smaller software packages for more flexible assembly and convenient secondary development. Moreover, it introduces abundant functional components, plugins, and enhances feature expansion for developers' convenience.
The algorithms receive a boost in the 9.0 platform, optimizing perception algorithms by integrating advanced models like the CenterPoint for LiDAR detection and the Yolo X + Yolo 3D model for visual enhancements. These models, trained on a massive dataset, exhibit significantly improved recall and precision rates. Incremental training, supporting independent model training, uplifts the detection capabilities for special targets and scenarios while maintaining the original detection capabilities at a lower cost.
Tools-wise, the 9.0 version introduces new functionalities like high-precision mapping, sensor calibration, and integration tools. The Dreamview+ debugging tool sees a comprehensive upgrade, enhancing multi-scenario usage, flexible layout, and resource utilization, streamlining debugging processes and window layouts for developers.
Morover, the platform's documentation undergoes a reconstruction for improved ease of operation, smoother reading experiences, and richer content, reducing the learning curve and usage costs for developers.
The Apollo Open Source Platform 9.0 demonstrates robust universal scenario capabilities, heightened user-friendliness, and scalability, reducing adaptation stages by 40%, code reading by 90%, and debugging by 80%, facilitating swift implementation of autonomous driving vehicles within a week, said Baidu. Additionally, its coverage of diverse scenarios enables completion of scenario application business systems within a month, notably shortening sensor calibration and map creation cycles to hourly levels, significantly expediting deployment efficiency.
The platform not only enhances software capabilities but also facilitates cost reduction and broader hardware compatibility. The expansion to configurations like 3 LiDAR + 4 Cameras and support for over four camera manufacturers and advanced devices such as GMSL, 32-line, and 64-line LiDAR from multiple brands signifies a significant upgrade in hardware versatility. Positioning devices now encompass support for over 3 major brands, facilitating comprehensive deployment in various scenarios.
