Apollo is an open-source autonomous driving platform developed by Baidu that provides a comprehensive software framework for building, testing, and deploying self-driving vehicles. Written primarily in C++, the platform addresses the full spectrum of autonomous vehicle development including perception, sensor fusion, localization, path planning, control algorithms, mapping, and simulation capabilities.
The repository demonstrates substantial ongoing development activity, with GitGenius tracking 353 issues and pull requests showing a median response latency of 29.4 hours, indicating active community engagement. The most frequently addressed areas are the Perception module with 27 tracked issues, the Simulation and Dreamview components with 19 issues, and the Cyber framework with 18 issues. The project's core contributors include daohu527 with 311 recorded events, yuxin-2 with 82 events, and YuqiHuai with 48 events, demonstrating concentrated expertise in maintaining the platform. The repository's contributor base overlaps significantly with major machine learning and computer vision projects including OpenCV, PyTorch, and TensorFlow, reflecting Apollo's reliance on these foundational technologies.
Apollo's architecture has evolved through nine major versions, each expanding autonomous driving capabilities. Version 1.0 introduced automatic GPS waypoint following for enclosed venues like test tracks. Version 1.5 added LiDAR-based perception for fixed lane cruising. Version 2.0 enabled autonomous driving on simple urban roads with collision avoidance and traffic light recognition. Version 2.5 introduced highway autonomous driving with camera-based obstacle detection. Version 3.0 and 3.5 progressively added support for complex urban scenarios including residential areas and unprotected turns. Version 5.0 focused on production-ready geo-fenced autonomous driving, while version 5.5 introduced curb-to-curb urban driving capabilities. Version 6.0 enhanced deep learning models and introduced data pipeline services. Version 7.0 added Apollo Studio as an online development platform alongside new perception and prediction models. Version 8.0 introduced a modular "Package" system for code organization and integrated model training, deployment, and validation tools. Version 9.0 continues this evolution with additional capabilities.
The platform requires specific hardware and software prerequisites. Vehicles must have complete by-wire systems including brake-by-wire, steering-by-wire, throttle-by-wire, and shift-by-wire capabilities, with testing performed on Lincoln MKZ vehicles. Development machines need at least an 8-core processor and 16GB memory, with NVIDIA Turing or AMD GFX9/RDNA/CDNA GPUs strongly recommended. As of November 2024, Apollo upgraded to CUDA 11.8 to support NVIDIA Ada Lovelace GPUs with driver version 520.61.05 or higher. The platform supports Ubuntu 18.04, 20.04, and 22.04, requires Docker-CE 19.03 and above, and the NVIDIA Container Toolkit for GPU support. The documentation recommends installing Apollo versions sequentially starting from 1.0 to ensure proper hardware validation before advancing to more capable versions.
