GAC Group: Enabling Intelligent,
Autonomous Driving with an AI Data Lake

As transportation advanced, cars started racing down roads faster than anyone could have previously imagined, but the evolution of mobility never stopped.

Today, transportation has taken a smarter turn. Powered by an AI data lake, the new generation of intelligent vehicles is accelerating toward the future of mobility.

GAC Group's Journey Towards Intelligent Mobility

GAC Group has been instrumental to the development of China's automotive industry.

Founded in the 1990s, it has ranked in the top 200 of the Fortune Global 500 for 12 consecutive years. When the digital era arrived, this major player in the Chinese automotive industry, based in South China, became a trendsetter.

As automakers pursue intelligent transformation, one of the most demanding challenges they face is undoubtedly realizing fully autonomous driving.

As the hub of the group's R&D system, the GAC R&D Center is responsible for exploring the application of cutting-edge technologies for their own brands such as Trumpchi, AION, and HYPTEC. The GAC R&D Center has pioneered the deployment of full-stack R&D for AI-powered autonomous driving, promoting the in-depth integration of key digital technologies, such as sensing and decision-making, with chassis-by-wire and power control. It has also driven the formation of a full-chain autonomous driving ecosystem, from R&D and production to operation.

The key to autonomous driving lies in precision and stability, both of which are highly dependent on mass data processing. Therefore, an advanced intelligent base is required to support GAC's R&D and verification, accelerating the path toward smarter autonomous driving.

Unlocking Data Value for Autonomous Driving

In a sense, all phases of autonomous driving, including development, testing, parameter fine-tuning, and model optimization, are inseparable from data.

Data storage, serving as the ultimate repository, bridges gaps and enables seamless operations across all phases.

Consider the first stage of data collection: a data collection vehicle has to brave the elements—scorching sun, biting cold, and rough, uneven roads—while the driver gathers rich, multidimensional environmental data, meteorological information, and extreme road conditions, ensuring that every possible scenario is captured.

For R&D personnel, handling the collected data is both challenging and rewarding. Valuable data from cameras, LiDAR, ultrasonic radar, and high-precision positioning systems must be fully preserved. This requires a storage cluster that is capable of strong horizontal scalability and flexible data management.

Furthermore, driven by large AI models, high-quality autonomous driving workflows involve over 10 phases, including data aggregation, compliance and anonymization, and cleansing and annotation. This means data repeatedly circulates across more than 10 development and testing applications. Under the pressure of tight model iteration cycles, huge demands are placed on the efficiency and management capabilities of the entire data infrastructure.

As a pioneer in the autonomous driving field, GAC had to address these pain points.

The sheer volume of data was staggering. Each data collection vehicle generated a massive 2 TB of raw data daily. After real-time preprocessing, 15% was extracted into scenario library data. Scaling up their fleet to just 100 road-test vehicles in the future would mean an estimated increase of over 1 PB of data per day. Even after cleansing and deduplication, they still project an astonishing 20 PB of new data in just six months.

But related pressure wasn't just about the data volume, and software-level challenges were equally critical. First, the pace of algorithm commercialization had been held back by complex development workflows. To break through this logjam, GAC needed to drastically improve data circulation and access efficiency and tear down protocol barriers. Second, as GAC's autonomous driving business embraced cloud transformation, the migration to containers and microservices placed even higher demands on the storage infrastructure. This led to far more stringent requirements for storage integration, permission isolation, and resource distribution, making a complete data foundation upgrade inevitable.

The bottom line? Traditional storage solutions simply couldn't keep pace.

AI Data Lake Solution: The Ark That Navigates the Data Ocean

Autonomous driving research is like swimming upstream—stop moving, and you fall behind. With this in mind, and looking to achieve a transformative leap in their development, GAC settled on adopting the AI data lake solution built on Huawei OceanStor Pacific scale-out storage. This move has effectively streamlined GAC's entire data loop, from raw data collection all the way to AI simulation.

Lake + Cloud

The AI data lake offers immediate benefits, starting with seamless integration. OceanStor Pacific scale-out storage smoothly connects with GAC's cloud, efficiently integrating with their data platform, container platform, and object gateway. Through its multi-tenant management feature, OceanStor Pacific enables more fine-grained and flexible control over resource provisioning, permission control, and metering services, resulting in simplified, powerful management.

Unified Services

Next comes the implantation of multi-protocol interworking. Powered by this key capability, Huawei's AI data lake solution enables one copy of data to be shared across multiple services without a need to replicate data. This is especially vital for upper-layer workflows using diverse standards and protocols at the preprocessing, training, and simulation phases. The solution implements interworking between NFS, CIFS, S3, and HDFS protocols without semantics loss or compromising performance.

Speeds and Savings

Finally, the solution delivers significant gains in terms of both performance and costs. Tailored and optimized for hybrid workloads, OceanStor Pacific boosts read/write efficiency by 40%, accelerating the entire data chain. Furthermore, the AI data lake solution recognizes data access frequency and smartly tiers data accordingly. This hugely streamlines data management and slashes TCO throughout the data lifecycle.

True to its name, OceanStor Pacific scale-out storage can accommodate an ocean of data, giving GAC peace of mind as they navigate a vast and ever-growing data lake—where every bit is harnessed, and every innovation finds its course.

Looking to the future, autonomous driving is only just beginning to transform human lives. GAC recently unveiled its upgraded intelligent driving and cockpit brand—ADiGO. This system is already backed by one billion kilometers of data training that covers 357 complex scenarios, including tidal lanes and traffic circles, and boasts an impressive 99% ramp pass rate.

Furthermore, GAC is set to launch China's first mass-produced Level 3 autonomous driving vehicle in the fourth quarter of this year.

From the busy flow of city traffic to the quiet corners of residential streets, GAC's journey represents a turning point for autonomous driving in China. Behind the scenes, Huawei Data Storage plays a vital role as the digital backbone, powering the vast data highways of a digital China with absolute reliability.

One day soon, as we relax inside our vehicles, gliding effortlessly through urban environments, we will remember the technological forces that quietly illuminated our daily life — turning dreams of smart, safe, and seamless travel into reality.