I. Safety Is the Core Proposition of Intelligent Driving

The safety and reliability of intelligent driving functions have become critical bottlenecks in industry development. Frequent incidents involving intelligent driving systems and growing public skepticism have compelled us, as automotive testing professionals, to confront a central question: In complex open-road environments, how can we establish a more scientific and comprehensive testing system to accurately identify system flaws and strengthen the safety defenses of intelligent driving?

II. Analysis of Intelligent Driving Functions and Testing Methods

Intelligent driving functionality has now become a highlight, a selling point, and even a standard feature in vehicles. Mainstream intelligent driving capabilities range from L2 to L2+ and even L2+++. Specific functions span from AEB and adaptive cruise control to highway NOA, memory parking, nationwide mapless NOA, valet parking, and autonomous driving. Sensor configurations have also evolved from 1R1V to 5R11V and further to 5R11V1L (or 5R12V3L). In February 2025, Tesla began rolling out FSD in China, intensifying competition and prompting major automakers to engage in a more aggressive technological arms race.

The rapid iteration of intelligent driving functions demands increasingly sophisticated testing solutions. Currently, testing for intelligent driving is primarily divided into simulation testing, closed-site testing, and open-road testing. Each approach has its strengths and limitations, which we have analyzed as follows:

Among these, open-road testing offers scenarios that most closely resemble real-world user conditions, with nearly 100% scenario overlap. This makes it the most effective method for uncovering potential issues in intelligent driving systems, though it is also the most challenging. Currently, automakers employ two main methods for open-road testing:

The first method involves using the vehicle’s domain controller or an onboard industrial computer to store vehicle data, mark points of interest, and upload relevant data. This primarily includes vehicle bus data, perception data, and control data. In simple terms, this approach relies on collecting and storing native vehicle data for self-inspection.

The second method involves installing a higher-precision testing system, including data acquisition systems, lidar, cameras, and integrated inertial navigation systems, complemented by truth post-processing systems. The stored data encompasses native vehicle bus data, perception and control data, as well as raw sensor data from the additional testing system, ground truth results, and points of interest. Simply put, this method uses superior reference data to detect and identify potential issues.

We have conducted a comparative analysis of both approaches.

III. Sharing Testing Experience and Solutions

How can we mitigate the limitations of open-road testing, close the loop on test data, use open-road testing to enhance simulation and closed-site testing, and simultaneously reduce testing complexity? Drawing on over 20 years of automotive testing experience, technological积累, and frontline market insights, we have focused on the following aspects:

1. ASEva Open-Road Data Closed-Loop Testing Solution

This solution provides an end-to-end data closed-loop testing framework, from vehicle to local to cloud platforms.

Capable of comprehensive, high-bandwidth multi-sensor data acquisition, it includes a ground truth system and supports bypass collection of data from native vehicle cameras, millimeter-wave radar, lidar, and domain controllers.

Enables large-scale, low-cost, high-quality dataset expansion based on road-collected data, supporting 3D/4D data annotation for algorithm model training.

Allows users to leverage bypass-stored native vehicle data for troubleshooting, helping testing teams quickly identify issues.

Supports scenario extraction, mining, and slicing, converting real-world collected scenarios into OpenX simulation scenarios to address the limited authenticity in simulation testing.

Facilitates vehicle perception and functional evaluation using ground truth data, generating extensive KPI metrics to optimize perception and intelligent driving algorithms.

Enables issue replication through data re-injection (partial or full), reducing the need for repeated road tests, shortening algorithm validation cycles, and mitigating the high safety risks of open-road testing.

Supplies key scenarios generated from open-road testing to closed-site testing, enriching test scenarios and alleviating the limitations of closed-site environments.

Offers cloud-based solutions addressing data management, vehicle monitoring, data mining, big data analysis, and third-party tool integration to enhance efficiency and shorten development cycles.

2. ASEva High-Precision Ground Truth Evaluation Solution

This solution equips test vehicles with higher-precision, wider-coverage, reduced-blind-zone sensors featuring perceptual redundancy. It generates ground truth based on vision-LiDAR fusion post-processing algorithms.

Provides superior perception capabilities compared to native vehicle systems, with higher accuracy and precision for evaluating perception systems, validating detection ranges, and identifying blind zones.

Evaluation targets include objects, lane lines, traffic signs, curbs, parking spaces, etc. Output metrics include detection rate, miss rate, false alarm rate, ranging accuracy, and velocity measurement accuracy.

3. rFpro + Ansible Motion High-Fidelity Simulation Testing Solution

rFpro is a high-precision driving simulation platform originally developed for F1 racing. It offers high-resolution digital twins of real public roads, test facilities, and proving grounds, effectively addressing insufficient model accuracy.

Applications include:

Generating high-precision comprehensive training datasets

Building high-fidelity sensor models

Automated and scalable high-fidelity scenario construction

End-to-end autonomous driving testing

Large-scale scenario testing and validation

In-loop simulation testing (DIL, VIL, HIL, SIL)

Intelligent headlight testing

Vehicle dynamics testing

Tire modeling testing

4. Additional Testing Solutions

To better meet user needs, we offer a range of other testing solutions:

ASEva Mini: A low-cost solution for generalized ADAS/AD function testing in open-road environments.

Track Range Testing System: For regulatory ADAS/AD testing in closed sites (also applicable for limited ground truth target testing in open roads).

ISAt Testing System: A solution specifically for Intelligent Speed Assistance (ISA) testing.

ASec Testing System: For testing ADAS/AD acoustic, optical, and vibration feedback.

SDT Testing System: For open-road ADAS/AD functional testing.

IV. Conclusion

We firmly believe that only by establishing a robust testing framework can we truly safeguard the reliable implementation of intelligent driving. If you have any questions, please do not hesitate to contact us for further details on testing solutions and equipment.