Key Infrastructure Considerations for Autonomous Driving: Storage, Computing, and Services

After attending a Xiaoma Zhixing offline sharing on autonomous driving, the author reflects on the essential infrastructure components required for self‑driving systems.

Storage : Autonomous vehicles generate terabytes of sensor data per day; three handling strategies are discussed – short‑term caching, local SSD storage with later upload, and hybrid network transmission combined with on‑board disk storage. A distributed file system is advocated for scalable, reliable storage of massive data volumes.

Database and Metadata : While traditional key‑value stores like HBase suit internet workloads, autonomous driving data differs in size and access patterns. The author argues that a simple file system plus rich metadata (capture time, location, weather, etc.) is sufficient for data management and retrieval.

Computing : Two computation layers are needed – offline batch processing (e.g., Spark for map generation, data cleaning) and online real‑time processing (CPU for general tasks, GPU for neural networks, FPGA for custom low‑latency algorithms).

Basic Services : To support higher‑level applications, services such as RPC, message queues, configuration centers, and container orchestration are required, similar to typical micro‑service architectures.

Real‑time Platform : The core of an autonomous driving platform reduces to a distributed message queue and a module scheduler, mirroring the design of Apollo Cyber. Efficient memory sharing (passing pointers) avoids costly data copies between perception and localization modules.

Business Scenarios : The author lists typical use cases – ride‑hailing, high‑precision map services, human‑machine interaction, vehicle fleet management, data collection pipelines, testing platforms, and simulation environments for validation and model training.

In summary, the article emphasizes that robust storage, metadata management, scalable computing, and well‑orchestrated services form the backbone of autonomous driving systems.