Driver assistance systems all need to know what is around the vehicle, but manufacturers disagree sharply about which sensors to fit. The disagreement is less about capability than about how failures should be covered.
What a camera can and cannot do
A camera produces a dense image with colour and texture, which makes it the only sensor that can read a road sign, distinguish lane markings or identify a traffic light.
What it does not produce directly is distance. Range must be inferred, either from two cameras viewing the same scene or from software trained to estimate depth from a single image.
Cameras also inherit the limits of vision. Low sun, heavy rain and darkness degrade them in the same ways they degrade a human driver.
Why radar covers the gaps
Radar measures distance and closing speed directly by timing a returned radio signal, and the frequencies used pass through rain, fog and spray with little loss.
That makes it dependable in exactly the conditions that trouble a camera, which is why radar has been standard on adaptive cruise control for years.
Its weakness is resolution. Radar can report that something is ahead at a given distance and speed, but it struggles to describe what that something is.
What lidar adds
Lidar sweeps laser pulses across the scene and times their return, producing a three-dimensional map with far finer detail than radar and with distance measured rather than inferred.
It works in darkness because it supplies its own light, and it gives precise shape information about objects the system has never encountered before.
The cost has historically been high and the units are physically awkward to package, though both problems have eased as solid-state designs have replaced rotating assemblies.
Why the industry split
One position holds that driving is a visual task solved by human eyes, so a camera system with sufficiently capable software should be enough, and additional sensors add cost and conflicting data.
The other holds that a system with no route to a fallback is unacceptable, and that overlapping sensors of different physical types are the only way to ensure a single failure mode does not blind the car.
Both positions are internally consistent, which is why the argument has run for years without resolution.
Where the difficulty actually sits
Combining sensors is not free. When a camera and a radar disagree about whether an object is present, something must decide which to believe, and that decision is where many assistance systems behave unpredictably.
Stationary objects are the classic case, because a radar return from a parked vehicle looks similar to one from a manhole cover or an overhead sign.
Adding a sensor therefore adds a fusion problem as well as a capability, and the quality of that fusion matters at least as much as the hardware list.