The automotive world is undergoing a seismic shift, moving away from traditional Electronic Control Unit (ECU)-based systems towards sophisticated software-defined vehicles (SDVs). This evolution promises greater flexibility, scalability, and modularity, but it also introduces complex challenges, particularly concerning real-time performance. For autonomous driving, where milliseconds can mean the difference between safety and risk, understanding and optimizing latency is paramount. A groundbreaking study from the Technical University of Munich sheds new light on how modern architectural approaches, like microservices and containerization, impact these critical real-time metrics.
:::info
Authors:
(1) Tobias Betz, Technical University of Munich, Germany;
(2) Long Wen, Technical University of Munich, Germany;
(3) Fengjunjie Pan, Technical University of Munich, Germany;
(4) Gemb Kaljavesi, Technical University of Munich, Germany;
(5) Alexander Zuepke, Technical University of Munich, Germany;
(6) Andrea Bastoni, Technical University of Munich, Germany;
(7) Marco Caccamo, Technical University of Munich, Germany;
(8) Alois Knoll, Technical University of Munich, Germany;
(9) Johannes Betz, Technical University of Munich, Germany.
:::
Table of Links
Abstract and I. Introduction
II. Related Work
III. Microservice Architecture for an Autonomous Driving Software
IV. Experiments
V. Results
VI. Discussion
VII. Conclusion, Acknowledgments, and References
The Evolution of Automotive Software: SDVs and the Need for Speed
Software-defined vehicles are revolutionizing how cars are designed and operated. At their core is a powerful central computing unit that orchestrates everything from sensor data processing to infotainment and advanced driver assistance systems. This architecture decouples software from hardware, fostering greater innovation and faster development cycles.
Central to this revolution are lightweight virtualization technologies, especially containers. These enable efficient resource utilization and strong isolation for software components, allowing applications to run independently in their own virtual environments. However, for critical applications like autonomous driving, stringent real-time criteria—such as maintaining low end-to-end latency and minimal communication jitter—must be met.
In autonomous vehicles, the timely processing of sensor data within strict deadlines is crucial for prompt decision-making and control. For instance, an end-to-end latency of 100 ms is often considered an acceptable benchmark for the entire chain, from sensor input to the vehicle’s trajectory controller. Failure to meet these real-time metrics can degrade performance and significantly increase the risk of accidents.
Microservices and Containerization: A Deep Dive into Latency
Despite the widespread adoption of containers in other industries, their specific impact on fundamental real-time metrics within complex automotive software has been largely unexplored. This gap is precisely what researchers from the Technical University of Munich aimed to address in their comprehensive study.
The study introduces a novel microservice architecture explicitly designed for Autoware, an open-source autonomous driving software built on ROS 2. This architecture was deployed on the research vehicle EDGAR. Critically, containers were used to isolate each service, providing a robust testing ground for real-world scenarios.
The researchers investigated the impact of containerization on end-to-end latency across various configurations, deploying the architecture on both x86 and aarch64 platforms. They used industry-standard tools like k3s for container orchestration and Docker for container management, ensuring the findings are relevant to current industry practices.
Surprising Results: Containers Outperform Bare Linux
Contrary to common assumptions that virtualization adds overhead, the study yielded surprising and highly significant results. The comprehensive evaluation showed that containerization can actually improve end-to-end latency in autonomous driving systems, even outperforming standard bare-Linux deployments.
Specifically, the microservice architecture demonstrated a mean end-to-end latency improvement of 5-8% when using container deployment. Furthermore, the maximum latencies—a critical concern for safety-critical systems—were significantly reduced. The study also evaluated communication jitter, system CPU, and memory utilization, all showing benefits from the containerized approach.
These findings challenge the conventional wisdom that bare-metal performance is always superior for real-time applications. They highlight the intricate nature of optimizing complex system scenarios and underscore the substantial benefits that containerization can bring to future SDV systems. The developed microservice architecture will be contributed open-source to the Autoware Foundation, fostering further innovation.
Conclusion: Paving the Way for Safer, Smarter Autonomous Vehicles
This groundbreaking real-world latency study of microservice architectures in autonomous driving provides invaluable insights for the future of software-defined vehicles. By demonstrating that containerization not only maintains but can actually improve crucial real-time metrics like end-to-end latency and reduce maximum latencies, the research opens new avenues for developing more robust, flexible, and ultimately safer autonomous systems.
The work by Tobias Betz and his team at the Technical University of Munich underlines the importance of rigorous, real-world testing to uncover performance characteristics that theoretical models or simpler benchmarks might miss. As the automotive industry continues its rapid transformation, embracing advanced architectural patterns like microservices and containerization, backed by such empirical evidence, will be key to unlocking the full potential of autonomous driving.
:::info
This paper is available on arxiv under CC by 4.0 Deed (Attribution 4.0 International) license.
:::
