GAC Group Builds an AI Data Lake on Huawei OceanStor Pacific for Autonomous Driving R&D

GAC Group's R&D Center, which develops technology for the group's Trumpchi, AION and HYPTEC brands, has built an AI data lake on Huawei OceanStor Pacific scale-out storage to support its autonomous driving research and verification, according to a Huawei case study. Every phase of autonomous driving, from development and testing to parameter tuning and model optimisation, depends on large volumes of data from cameras, LiDAR, ultrasonic radar and high-precision positioning systems. The data volumes are large: each data collection vehicle generates 2 TB of raw data a day, of which 15% is extracted into scenario library data after real-time preprocessing. Expanding to 100 road-test vehicles would add an estimated more than 1 PB of data a day, and even after cleansing and deduplication GAC projects 20 PB of new data in six months. Workflows driven by large AI models involve more than 10 phases, including data aggregation, compliance and anonymisation, and cleansing and annotation, so data circulates through more than 10 development and testing applications. GAC also needed storage that fitted its move to containers and microservices, with stricter permission isolation and resource allocation. The OceanStor Pacific data lake connects with GAC's cloud, data platform, container platform and object gateway, and its multi-tenant management gives fine-grained control over resource provisioning, permissions and metering. Multi-protocol interworking across NFS, CIFS, S3 and HDFS lets one copy of data be shared by preprocessing, training and simulation without replication. The system tiers data automatically by access frequency to lower total cost of ownership, and Huawei says OceanStor Pacific raises read/write efficiency by 40%, streamlining GAC's data loop from raw collection to AI simulation. Huawei notes that GAC's upgraded intelligent driving and cockpit brand, ADiGO, has been trained on one billion kilometres of data covering 357 complex scenarios, with a 99% ramp pass rate.