Automotive autonomous driving function testing verifies the perception, decision-making, and control performance of advanced driver assistance systems (ADAS) and higher-level automated driving functions under controlled, repeatable, and standards-based conditions. This article outlines the technical framework, sample types, testing items, evaluation methodologies, and typical applications of autonomous driving function testing performed by accredited third-party laboratories, with reference to international and national standards such as ISO 26262, ISO 21448, ISO 34502, and the GB/T series.
Technical Overview
Automotive autonomous driving function testing is a systematic verification and validation process that evaluates whether a vehicle's automated driving system can correctly perceive its environment, make safe and appropriate driving decisions, and execute precise control actions across its declared Operational Design Domain (ODD). Driving automation is commonly classified into levels from L0 (driver assistance) through L5 (full automation), and the depth of testing increases significantly as the level of automation and the transfer of driving responsibility from the human to the system increase.
Unlike conventional automotive component testing, autonomous driving function testing addresses a complex cyber-physical system in which sensors, fusion algorithms, planning software, and by-wire actuators interact with a highly dynamic and uncertain traffic environment. Consequently, the evaluation framework combines several complementary standards families: ISO 26262 for functional safety of electrical and electronic systems, ISO 21448 for Safety of the Intended Functionality (SOTIF), ISO 21434 for cybersecurity engineering, and product-level performance standards such as ISO 15622 for adaptive cruise control, ISO 11270 for lane keeping assistance, and ISO 22839 for forward vehicle collision mitigation. In the Chinese regulatory context, GB/T 40429 defines the driving automation classification, while GB/T 39263 and GB/T 39901 specify test methods for ADAS functions.
The overall objective is to demonstrate, with quantified evidence, that the system performs its intended function correctly in known scenarios, behaves safely in unknown or degraded conditions, executes a minimal risk maneuver when the function reaches its limits, and achieves an acceptable residual risk level before market release or regulatory homologation.
Testing Samples
Complete Vehicles and System Prototypes
Passenger cars, robotaxis, automated shuttles, and commercial trucks equipped with production-intent or pre-production automated driving systems are tested as integrated wholes. Complete vehicle testing evaluates the end-to-end behavior of the sensor set, computing platform, actuation chain, and human-machine interface under real dynamic conditions.
Domain Controllers and Electronic Control Units
ADAS/AD domain controllers, brake and steering controllers, and gateway units are tested on the bench to verify computing performance, interface consistency, diagnostic behavior, timing constraints, and fail-safe responses before integration into the vehicle.
Sensor Modules
Individual perception sensors, including automotive cameras, 4D millimeter-wave radar, LiDAR, ultrasonic sensors, and GNSS/IMU positioning modules, are characterized for detection range, angular coverage, resolution, accuracy, and robustness against environmental interference.
Software Stacks and Perception Algorithms
Perception, sensor fusion, prediction, and planning software releases are evaluated using recorded data sets, replayed scenarios, and virtual environments to quantify detection rates, classification accuracy, and decision quality before deployment on target hardware.
V2X and Communication Units
Onboard units for vehicle-to-everything (V2X) communication, high-definition map modules, and data recording devices are tested for protocol conformance, latency, positioning accuracy, and compliance with applicable GB and IEC requirements.
Testing Items
Perception Performance
- Detection range, field of view, and target resolution for vehicles, pedestrians, cyclists, and static obstacles
- Classification accuracy, tracking stability, and false positive or false negative rates
- Performance under adverse weather, low illumination, glare, fog, and soiled sensor surfaces
- Positioning accuracy and availability of GNSS/IMU and map-matching localization
Function-Level Performance
- Adaptive cruise control: speed tracking accuracy, headway keeping, cut-in response, and smooth stop-and-go behavior per ISO 15622
- Automatic emergency braking: collision avoidance and mitigation rates across vehicle, pedestrian, and cyclist scenarios
- Lane keeping and lane centering: lateral deviation, steering smoothness, and curve negotiation per ISO 11270
- Blind spot detection, lane change assistance, traffic jam assist, and highway pilot functions
- Automated parking: slot detection accuracy, trajectory precision, and obstacle response
Decision, Planning, and Control Quality
- Obstacle avoidance path planning and dynamic replanning capability in complex interactions
- Longitudinal and lateral control accuracy, stability, and passenger comfort metrics
- Takeover request design and driver response measurement during transitions of control
- Minimal risk maneuver (MRM) execution when the system exits its ODD or experiences degradation
Safety Assurance and Compliance
- Functional safety verification per ISO 26262, including fault injection and ASIL-rated safety mechanisms
- SOTIF analysis and validation per ISO 21448, addressing performance limitations and unforeseeable misuse
- Cybersecurity testing per ISO 21434, including penetration resistance of communication and update interfaces
- Compliance with driver monitoring, event data recorder, and regulatory notification requirements
Environmental and Electrical Robustness
- Electromagnetic compatibility of sensors and controllers under IEC-based automotive EMC test methods
- Climatic, mechanical vibration, and ingress protection (IP) verification for sensor housings and control units
- Power supply robustness, including voltage transients and load dump conditions
Testing Methods
Virtual and Simulation-Based Testing
Model-in-the-loop (MIL) and software-in-the-loop (SIL) simulations evaluate perception, planning, and control algorithms against large-scale scenario libraries. Scenario-based safety evaluation frameworks aligned with ISO 34502, combined with standardized road network and scenario description formats, enable coverage-driven testing of millions of kilometers in compressed time and statistical extrapolation of residual risk.
Hardware-in-the-Loop and Vehicle-in-the-Loop Testing
Hardware-in-the-loop (HIL) benches connect real controllers to simulated vehicle dynamics and virtual sensor streams, allowing verification of timing, fault handling, and diagnostic behavior under reproducible edge cases. Vehicle-in-the-loop (VIL) methods place the physical vehicle on a proving ground while it interacts with augmented virtual traffic, bridging simulation and real-world testing.
Closed-Course Proving Ground Testing
Physical track testing uses instrumented vehicles, crashable soft target platforms, articulated pedestrian and cyclist dummies, and robotic driving robots to reproduce standardized scenarios such as car-to-car rear, crossing, and cut-in conflicts. Ground truth is captured with high-accuracy RTK GNSS/IMU reference systems, mobile mapping, and synchronized multi-channel data acquisition, allowing distance, speed, time-to-collision, and control intervention quality to be measured with centimeter-level precision.
Public Road and Long-Duration Testing
Controlled open-road evaluations confirm real-world performance, availability, and user interaction across diverse traffic densities, road types, and weather conditions. Long-duration fleet testing accumulates statistical evidence on disengagement rates, false activation frequency, and system availability in support of release decisions and regulatory submissions.
Fault Injection and Robustness Testing
Deliberate sensor occlusion, signal degradation, communication loss, and actuator faults are introduced to verify degradation detection, safe state transitions, and fail-operational behavior, satisfying both ISO 26262 fault coverage requirements and SOTIF insufficiency analysis.
Applications
Autonomous driving function testing supports a wide range of engineering, regulatory, and commercial objectives across the vehicle lifecycle:
- Regulatory homologation and type testing for market access in accordance with applicable national regulations and GB standards
- Design verification and validation during development, including release sign-off before start of production
- Supplier acceptance testing for domain controllers, sensors, and software releases
- Safety case construction and residual risk argumentation for insurers, fleet operators, and authorities
- Deployment approval for automated shuttles, robotaxi services, autonomous freight, and closed-site logistics such as ports and mines
- Post-market surveillance, incident reconstruction, and over-the-air update revalidation
FAQ
What is the difference between ADAS testing and autonomous driving function testing?
ADAS testing focuses on individual driver assistance functions such as AEB or ACC, where the human driver remains responsible and must supervise continuously. Autonomous driving function testing additionally covers the complete perception-decision-control chain, ODD boundary management, transitions of control, minimal risk maneuvers, and higher-level safety assurance under ISO 26262 and ISO 21448, because the system itself carries part or all of the driving task.
Why is SOTIF testing necessary when functional safety is already covered?
ISO 26262 addresses risks caused by system faults, while ISO 21448 addresses hazards arising from functional insufficiencies without a component failure, such as sensor performance limits, algorithmic misjudgment, or foreseeable misuse. Both dimensions must be tested because a fault-free automated driving system can still cause a hazard if its perception or decision logic is inadequate in certain conditions.
Can simulation completely replace real-world testing?
No. Simulation provides scalable scenario coverage, statistical evidence, and safe evaluation of dangerous edge cases, but physical proving ground and road testing remain mandatory to validate sensor characteristics, actuator behavior, environmental effects, and human interaction with real hardware. Current best practice combines scenario-based virtual testing with a defined portfolio of track and road tests agreed with accredited third-party laboratories and authorities.
How are test scenarios selected?
Scenarios are derived from regulatory test catalogs, standards such as ISO 34502, naturalistic driving data, accident databases, and risk-based analysis of the declared ODD. A coverage metric ensures that critical scenario categories, parameter boundaries, and combination effects are exercised, and unknown-unsafe scenarios are progressively converted into known-safe coverage through iterative SOTIF evaluation.
What deliverables are provided after testing?
Typical deliverables include a test plan with pass or fail criteria, raw measurement data with ground truth traces, quantitative performance and safety metrics, fault injection results, scenario coverage analysis, and a formal test report that can be used for internal release decisions, supplier acceptance, or submission to regulatory authorities.