What Driver-Assistance Sensors Cannot See
Driver-assistance systems are described by what they can do. The more useful question for anyone relying on one is what their sensors cannot see.
Three sensor types, three blind spots
Cameras see the way we do — colour, lane markings, signs, text — and fail the way we do. Low sun, heavy rain, fog, snow on the lens, and the transition into and out of tunnels all degrade them, and they judge distance less directly than the other two.
Radar measures distance and closing speed well and works through weather that blinds a camera. It has poor resolution, historically struggles with stationary objects, and cannot read a sign or a lane line.
Lidar builds precise three-dimensional geometry and works in the dark. It is degraded by heavy precipitation and by spray, and it does not read colour or text either — it knows an object's shape and position, not what it means.
Fusing them covers more cases than any one alone, which is the design intent. It does not produce a sensor set with no limits.
The cases that remain hard
Consistent across systems: stationary objects in the lane at speed, unusual shapes and orientations, poor or absent lane markings, roadworks with temporary layouts, heavy spray, low sun directly ahead, and anything the system's training and design did not anticipate.
That last category is the important one, and it is why the operational design domain matters more than the sensor list. A system is defined for conditions; outside them it is not degraded so much as inapplicable.
Why this connects to the SAE level
At Level 2 you are the fallback for exactly these cases. The system handles the ordinary situation well and hands you the unusual one — often with little warning, because a system that could reliably predict its own failure would not have the failure.
Which is the practical argument against treating a good Level 2 system as more than it is. Its competence in the common case is not evidence about the uncommon one, and the uncommon one is what you are there for.
The physics does not change
Whatever the sensors detect, stopping still takes the distance it takes. At 100 km/h on a dry road that is 97.85 m with a 1.5-second reaction, and 139.99 m in the wet — and an assistance system braking sooner helps precisely because it removes reaction time, not because it changes the braking.
Following distance is therefore the thing that makes any of it work. A system with a short following distance has the same braking physics as a driver with one, and both need the road ahead to be long enough.
The honest position
These systems are genuinely useful and measurably reduce some categories of collision. They are also specified for conditions, limited by sensors, and — below Level 3 — dependent on a supervising driver.
Knowing which level your vehicle is, what its domain is, and what it is documented not to handle is the difference between using a system and trusting one. The vehicle's own documentation is the authority on all three.
Keeping the sensors working
The practical maintenance is unglamorous and matters: keeping cameras, radar housings and sensor apertures clean, since mud, snow, ice and even a bad wash can degrade or disable a system without much warning.
Most vehicles will report a blocked sensor, and the correct response is to treat the assistance as unavailable rather than to assume it is working at reduced capability. Windscreen replacement often requires camera recalibration for the same reason.
What the systems are genuinely good at
Worth ending on, because the caveats can read as dismissal. Automatic emergency braking has measurable evidence behind it for reducing rear-end collisions, and blind-spot monitoring and lane-keeping address well-understood failure modes in human attention.
The argument is not that assistance is unreliable — it is that knowing its domain is part of using it. A system used inside what it was designed for is a genuine safety gain.