Event Recap | Product Director Lu Xiwen of Carlinx, Unpacks the Path to High-Value Data Closed-Loop Implementation

“World Models,” “Reinforcement Learning,” “Model Self-Evolution” – yet a more fundamental challenge remains unsolved: Where does the data driving these advanced paradigms come from? And how to unlock high-value data?

The 13th Conference on Intelligent Connected Vehicle Technology (CICV 2026) was recently held in Shanghai. Lu Xiwen, Product Director of Carlinx, was invited to deliver a keynote speech titled “High-Value Data Application Practices for Intelligent Connected Vehicles: From Data Acquisition to AI Closed Loop.” He shared a collaborative data application framework spanning multi-source data acquisition and governance to high-level scenario filtering and construction, ultimately enabling an AI-driven reasoning closed loop.

The following is an excerpt from his speech:

Industry Inflection Point: Data Quality Defines the Ceiling of AI Capability

Intelligent connected vehicles are accelerating into the “era of large models.” Whether it is the training and iteration of end-to-end autonomous driving models, scenario optimization of intelligent cockpit large models, or AI-driven diagnostic upgrades for vehicle health management, the core driving force points to the same element: high-quality vehicle-side data.

Mr. Lu pointed out that the industry faces a structural contradiction – “data explosion but value scarcity.” The number of ECUs continues to grow, sensor signal density keeps rising, and data volume expands exponentially. Yet the truly high-value data capable of supporting model training, fault diagnosis, and scenario generalization remains severely insufficient. “Data quality defines the ceiling of AI capability,” he emphasized. “Without a continuous input of high-quality data, the model intelligence has nothing to draw from.”

Technical Breakthrough: Building a Millisecond‑Precision Data Highway

Moving from “full‑scale acquisition” to “precision acquisition,” vehicle‑cloud collaboration is the key. To meet the demand for high‑value data in the AI era, Carlinx proposes an integrated approach of “precision acquisition + intelligent closed loop,” constructing a data collaboration system that covers the vehicle side, cloud side, and AI training pipeline.

On the vehicle acquisition layer, Carlinx’s self‑developed flexible data acquisition platform supports fine-grained acquisition at 200ms intervals. For key events such as minor collisions, hard acceleration, battery anomalies, and abrupt lane changes, it can fully retain multi‑domain correlated data before and after the event. Instead of the traditional “full upload to cloud” model, Carlinx emphasizes “collecting truly valuable data at the right time.” Through event triggering, edge computing, and dynamic task mechanisms, the system adapts acquisition strategies in real time based on scenarios – reducing bandwidth and storage costs while increasing the proportion of effective data. The platform also supports real‑time correlated upload of cross‑domain data from chassis, powertrain, cockpit, and autonomous driving, solving the “data silo” problem of traditional vehicle systems and providing a unified data foundation for subsequent AI analysis and scenario reconstruction.

On the transmission layer, high‑compression‑ratio data encoding technology further reduces the cost of uploading data from large fleets to the cloud, making long‑term, continuous data operations truly commercially viable.

The core of the AI closed loop is not just training models – it’s enabling data to evolve continuously.

On the cloud analysis layer, Carlinx has built a complete closed‑loop system covering data acquisition, scenario recognition, anomaly attribution, and model optimization. After multi‑dimensional data uploaded from vehicles enters the cloud, the system automatically performs scenario recognition and anomaly analysis. Based on the analysis results, it generates new data strategies and training requirements, which are then sent back to vehicles in real time. This means: when a certain type of anomalous scenario is identified, the system can automatically expand the relevant data collection scope, continuously supplement training samples, and drive model improvement. This closed loop of “acquisition – analysis – training – re‑acquisition” allows vehicle intelligence to evolve continuously in real‑world driving conditions, rather than remaining stuck in laboratory data phases.

Scenario Implementation: Full‑Scenario Value Release of Data + AI: From “experience‑based diagnostics” to “data‑driven diagnostics.”

In the intelligent diagnostics scenario, Carlinx is deeply integrating AI capabilities with the vehicle‑cloud data closed loop to build a new generation of remote intelligent diagnostics system. Traditional diagnostics rely heavily on manual experience and single fault codes. When facing complex faults, intermittent issues, or cross‑domain anomalies, they often suffer from long localization cycles and difficulty in reproduction. Leveraging multi‑source data correlation, knowledge graphs, and AI analysis, Carlinx can automatically correlate fault scenarios, intelligently match historical cases, jointly analyze multi‑domain signals, and generate remote localization suggestions and diagnostic recommendations. In some complex fault scenarios, the system has achieved diagnostic accuracy exceeding 90%, while significantly improving efficiency for routine issues – helping automakers shorten after‑sales cycles and reduce vehicle downtime.

For global operations, data capability is becoming the new infrastructure for Chinese automakers. Carlinx’s data acquisition and analysis platform has already been validated across multiple vehicle models in strictly regulated European as well as Southeast Asian markets. Facing different network environments, regulatory requirements, and data compliance frameworks, the platform stably supports cross‑regional data collaboration and remote operations, providing a unified data foundation for the globalization of Chinese automakers.

As the automotive industry moves into the era of “software‑defined vehicles” and “AI‑driven vehicles,” competition is shifting from single functional capabilities toward data capabilities and continuous evolution. The data closed loop is becoming the core infrastructure for the next phase of smart vehicle competition.