EN / 中文

Can Autonomous Driving Systems Tell When Sensor Data Is No Longer Trustworthy?

by zhijiazuiqianyan·November 16, 2026

Some time ago, a vehicle operating in assisted driving mode was entangled by a kite string on an elevated road, resulting in damage to the roof-mounted LiDAR.

Although this incident occurred some time ago, revisiting industry developments reveals that intelligent driving has long focused on enabling vehicles to see farther and more clearly.

Camera resolutions have steadily improved, LiDAR detection capabilities have continuously advanced, and the number of sensors on vehicles has increased. However, as more sensors are directly exposed to real-world road environments, another issue arises: beyond enabling cars to perceive the outside world, should we also ensure these sensors themselves remain functional?

This is not a new problem, but it has become increasingly critical as intelligent driving systems grow more dependent on external sensors.

01. Why Sensor Protection Is Becoming Increasingly Important

To acquire comprehensive environmental information, cameras, LiDARs, and millimeter-wave radars are positioned on the roof, bumpers, fenders, and other locations.

This layout expands the vehicle's perception range but also exposes more critical hardware to complex environments involving dust, mud, ice, snow, insects, and impacts from foreign objects.

Sensor performance parameters can be compared under controlled test conditions, but real roads do not always provide ideal working environments.

No matter how far a LiDAR can detect, its actual perception capability may degrade if the protective window is contaminated; no matter how high a camera's resolution, the information it provides is compromised if the lens is obscured by mud or water.

Therefore, behind the kite string incident, the true issue worth discussing is not whether a thin string can be detected by LiDAR, but how intelligent driving systems handle potential occlusion, contamination, and damage to sensors themselves, given their increasing reliance on external sensing.

Previously, the industry focused on making sensors see farther and clearer. As vehicles operate in more complex real-world environments, ensuring long-term sensor stability becomes a critical challenge for perception systems.

02. Can Data Be Trusted If It Isn't Broken?

For autonomous driving systems, a complete sensor failure is not necessarily the most complex scenario.

Device disconnection or data interruption is relatively easy to detect.

Mass-produced vehicles possess certain sensor fault diagnosis capabilities. When the system identifies significant anomalies in a sensor, it can adjust related functions based on the remaining capabilities of other sensors and restrict intelligent driving features if necessary.

The more difficult situation occurs when a sensor remains operational but fails to provide fully reliable data. For example, if a camera lens is partially obstructed, images may still be output normally; if a LiDAR's protective window is contaminated, it may continue generating point clouds.

From a system perspective, the device remains online, but the quality and integrity of the data may have changed. This highlights an often-overlooked aspect of sensor reliability: whether a device is online is distinct from whether its data remains trustworthy.

Current intelligent driving systems monitor sensor status through various methods. However, determining whether data quality has been affected by partial occlusion, local contamination, or gradual degradation in perception capability is often harder than detecting a complete device failure.

Thus, intelligent driving systems must address not only what happens ahead but also whether the information currently being received is still credible.

These are two different directions: one concerns whether the car can correctly understand its surroundings, while the other concerns whether the system can timely detect changes in its own perception capabilities.

03. Does More Sensors Mean Safer Driving?

Increasing the number of sensors is one of the most direct ways to mitigate issues with individual sensors.

If cameras are affected, LiDARs remain available; if LiDARs malfunction, millimeter-wave radars can provide supplementary information.

However, multi-sensor redundancy is not simply about adding backups.

Cameras, LiDARs, and millimeter-wave radars gather different types of information. Cameras excel at providing color, texture, and semantic details; LiDARs directly capture the spatial structure of the environment; and millimeter-wave radars offer advantages in distance and speed measurement.

Under normal conditions, they complement each other. But when results from different sensors contradict, the system must determine whether the discrepancy stems from the environment itself or from a change in a specific sensor's status.

For instance, if a camera fails to identify a target ahead while a LiDAR detects an anomalous object, it could indicate that the camera's field of view is obstructed or that the LiDAR is producing erroneous echoes.

Inconsistencies between different sensors cannot be resolved simply through cross-validation, as multiple sensors may be simultaneously affected in conditions such as rain, snow, or fog.

Therefore, the complexity of multi-sensor fusion lies not just in aggregating more data, but in continuously evaluating discrepancies between different information sources to assess which data remains reliable.

Adding sensors increases redundancy but also requires the system to manage greater uncertainty.

04. How Should Vehicles Drive When Perception Is Impaired?

Detecting a decline in perception capability is only the first step in the safety protocol for intelligent driving.

More importantly, the system must decide how to proceed after judging that its perception capabilities have changed.

From a system design perspective, responses to such issues are not limited to maintaining current functionality or immediately exiting automated mode.

A temporarily affected side-view camera differs significantly from simultaneous anomalies in multiple key sensors regarding impact on driving capability. Therefore, the system should adopt tiered response strategies.

Intelligent driving systems can adjust driving strategies based on remaining perception capabilities.

For example, should the vehicle reduce speed when perception in certain areas degrades? Should it minimize higher-risk maneuvers like lane changes when information from a specific direction is insufficient? Under what conditions should the system limit functionality or request driver takeover when it can no longer reliably confirm the surrounding environment?

These questions ultimately point to the same capability: the system's ability to redefine the current driving envelope based on its own state.

In the past, intelligent driving discussions largely focused on how vehicles make correct decisions in complex road scenarios. As sensor suites and system architectures become more complex, the ability to adjust driving behavior promptly when environmental perception capabilities change is becoming increasingly vital.

05. Final Thoughts

From cameras to LiDARs, competition in intelligent driving over the past few years has largely centered on sensor performance, with emphasis on detection range, resolution, and the ability to identify small targets.

However, once entering complex road environments, even the strongest single-sensor performance cannot guarantee that the entire perception system remains in optimal condition.

Obstruction or damage to a single sensor may not be alarming in itself; what matters more is whether the system can timely detect such changes and adjust subsequent strategies based on remaining perception capabilities.

Consequently, future competition in intelligent driving may not solely revolve around whose sensors perform better, but also around who can more effectively manage the capability boundaries of the entire perception system.