AVCS: Analysis of the Modular Open-Source Autonomous Driving Control System Architecture

AVCS is a modular open-source autonomous driving control system architecture that covers four core modules: perception, localization, planning, and control. It achieves module collaboration through ROS2 middleware and the hardware abstraction layer. This article will analyze its architectural design, core module functions, simulation test support, technology stack, and open-source value, providing references for developers and researchers.

Autonomous driving technology is moving from the laboratory to large-scale commercial use, and a reliable control system is the core foundation. As an open-source autonomous driving control framework, AVCS integrates hardware, software, sensors, artificial intelligence, robotics, and communication technologies. It supports autonomous vehicle driving, navigation, obstacle detection, and driving decision-making, providing an integrated technical platform for developers.

AVCS adopts a highly modular layered architecture, divided into perception, localization, planning, control, and human-machine interaction modules. It communicates and coordinates through ROS2 middleware, and the bottom layer interacts with sensors/actuators via the Hardware Abstraction Layer (HAL). The modular design achieves module decoupling—each module focuses on its own function (e.g., perception does not need to care about control execution), improving system maintainability and scalability, and facilitating customized development for different vehicle models and scenarios.

The perception module uses multi-sensor fusion (lidar, camera, millimeter-wave radar) to compensate for the limitations of a single sensor, and identifies road elements through target detection and tracking algorithms. The localization module adopts multi-source fusion (GPS, IMU, lidar, visual odometry), combined with Kalman filtering/particle filtering to achieve centimeter-level positioning accuracy, solving problems such as GPS signal occlusion and IMU cumulative errors.

The planning module includes global path planning (based on high-precision maps and navigation, considering road topology and traffic rules) and local path planning (real-time obstacle avoidance adjustment), using algorithms such as RRT* to generate smooth trajectories. The control module uses Model Predictive Control (MPC) and pure pursuit control to coordinate longitudinal and lateral control, ensuring the comfort and stability of vehicle trajectory tracking.

AVCS integrates simulators such as CARLA, Gazebo, and LGSVL, supporting sensor simulation, vehicle dynamics simulation, and traffic flow modeling. Developers can safely test extreme scenarios (bad weather, sensor failures), and large-scale parallel testing accelerates algorithm iteration. The modular design allows seamless migration of simulation code to real vehicles, shortening the deployment cycle.

AVCS is developed using C++17 (for underlying algorithms and real-time modules) and Python 3.10+ (for upper-layer logic and prototype development). It relies on the ROS2 Humble Hawksbill middleware, supports CUDA 11.8+ GPU acceleration, and provides a Docker containerized deployment solution to simplify environment configuration and cross-platform migration.

The AVCS open-source project contributes to the popularization and education of autonomous driving. Academically, it provides a standardized experimental platform, allowing researchers to focus on algorithm improvement. Industrially, it enables rapid development of scenario-specific solutions based on the modular architecture. The open-source system promotes technology iteration, talent cultivation, and ecosystem building, facilitating the open and transparent development of the industry.