The automotive industry stands at the threshold of a new era, one defined by a transformative leap from electrification to intelligent systems that prioritize advanced technologies, particularly those infused with artificial intelligenceZhang Yongwei, the Vice Chairman and Secretary-General of the China Electric Vehicle Hundred Association, articulated a vision of this shift at the recent "Dajun Mountain Intelligent Automotive Technology Conference (2024)." He painted a picture of an automotive landscape where, within a mere one to two years, the concept of a non-intelligent vehicle may vanish entirely, given the rapid pace of innovation.

As we delve into the statistics provided by the Ministry of Industry and Information Technology, we see a clear movement towards this intelligent realmIn the first half of this year, the penetration rate of level 2 and above advanced driver-assistance systems in new passenger cars reached an impressive 55.7%, with vehicles equipped with Navigation On Autopilot feature attaining a rate of 11%. Projections indicate that smart connected vehicle sales will surpass 17 million units by the year’s end, reflecting an unprecedented 60% penetration rate

With car manufacturers ramping up investments in intelligent technology, the race to develop higher levels of autonomous driving and large model applications is accelerating.

However, the journey towards fully autonomous driving is fraught with challenges, particularly regarding the need for data — a critical component in refining driving modelsAs automakers gear up for this digital age, the collection and processing of data remain significant hurdlesIn China, automotive companies face obstacles such as limited data collection scale, underrepresentation of long-tail scenarios, high costs, and inefficiencies in data acquisitionIndustry experts are suggesting that data sharing could be the key to overcoming these challenges.

The sharing of data has the potential to not only reduce collection costs, but also to significantly enhance the frequency of model training iterations

With the collapse of traditional boundaries, the paradigm of smart driving is increasingly shifting from a rules-based system to one driven by data — a transformation compelled by the recent surge in artificial intelligence and large model technologiesConsequently, high-quality data emerges as an essential ingredient for effective model training.

Li Bin, CTO of Beijing Kaimeng Data Technology Co., indicated that Tesla’s ability to amass over two million vehicles worldwide to collect data across diverse scenarios at significantly lower costs is a benchmark for othersIn contrast, the limited number of vehicles capable of such extensive data collection in China means that automakers can only gather data in specified regions, often hampered by stringent qualifications needed for data acquisitionAs such, achieving a critical scale of autonomous driving data in China presents a formidable challenge.

The demand for data is ballooning as the automakers scale their operations and capabilities; however, this skyrocketing appetite for high-quality data coincides with rising costs, creating a paradox that necessitates innovative solutions

For instance, Li elucidated a scenario where a requirement of 400,000 frames of data would roughly take 40 weeks to collect and an additional 80 weeks to process, costing over 5 million yuanBut through a cooperative data-sharing model, where some data is independently collected by companies while others are derived from suppliers or different manufacturers, the duration for both collection and processing could be halved, thereby reducing expenses considerably.

This hypothetical data circulation model not only promises cost efficiency but also the potential for improved training frequencies in autonomous driving models — leading to superior outcomes for automotive companiesCurrently, the prevailing attitude among automakers is one of hesitancy to share data, viewing it as a core asset, which exacerbates the prevailing issue of "data islands" within the industryHowever, the diverse nature of the scenarios required for training sophisticated autonomous systems highlights the necessity for collaboration

alefox

By embracing data sharing, the industry can mitigate redundancy in collection and processing, expand sample sizes, and enhance model performance, especially in extreme scenarios.

Experts in the field emphasize that fostering a culture of data sharing is vital for the maturation of smart driving technologies and facilitating their large-scale application on a global scaleThe consensus is clear: breaking down barriers to data access and resolving bottlenecks in computational capabilities is crucial for the industry as a whole.

During the conference, Liu Bo, Vice President and General Manager of the Automotive Transportation Division at Shanghai Zero Data Technology, elucidated the idea that data generated by different actors over time can only contribute effectively to growth when freely sharedHe contrasts the previous model of internal data circulation — which captured efficiencies within companies — with the necessity of external circulation that crosses organizational and industry boundaries

Liu envisions a collective effort to establish a data interoperability platform in the automotive sector as foundational to addressing key challenges — issues such as data modification, ownership disputes, uncontrollable flows of information, and safeguarding against privacy breaches.

On a similar note, Huo Jingyu, General Manager of East Division at Four Dimensional Map New Technology Co., discussed the exponential growth in data generated by smart cars, estimating that each vehicle produces over 150 parameters every few seconds, accumulating to daily data collections reaching nearly 100 TBYet, the enduring hesitation among automakers to utilize sensitive driving data presents formidable barriers, compounded by lengthy and complex data processing chains.

In light of these challenges, the emergence of a robust AI infrastructure is seen as a promising solutionHuo highlighted the importance of developing a cohesive system that outlines which data can be collected, methodologies for processing, data storage protocols, and how to ensure privacy is maintained

This comprehensive, closed-loop handling of data is essential for extracting the valuable insights required for automated driving trainingThe willingness of Four Dimensional Map New Technology to offer their algorithms for open collaboration with automakers exemplifies the cooperative spirit necessary to realize this vision.

Furthermore, Kaimeng Data’s ambition to serve as a connector promoting fluid data movement is reflected in their partnership with Beijing Yizhuang Intelligent City Research Institute to launch the nation’s first "Vehicle-Road-Cloud Data Collaboration Platform." This initiative focuses on efficiently and securely harnessing intelligent data application services that meet the needs of autonomous vehicle manufacturers.

Ultimately, facilitating the flow of automotive data can dismantle the existing silos that restrict the autonomous driving field