How LiDAR, Radar, and Cameras Work Together in Self-Driving Cars

Ask what a self-driving car "sees" and the honest answer is that it doesn't rely on any single sense the way a human driver relies on eyesight. Instead, autonomous vehicle stacks combine several fundamentally different sensor types, each strong where the others are weak, and blend their outputs into a single model of the world — a process engineers call sensor fusion. Understanding what each sensor actually measures explains why the combination exists at all.

LiDAR: precise shape, vulnerable to weather

LiDAR — short for Light Detection and Ranging — works by firing rapid pulses of laser light in every direction and timing how long each pulse takes to bounce back off a surface and return to the sensor. Because light travels at a known, constant speed, that round-trip time converts directly into a precise distance measurement. Spin the laser continuously and you build up a dense 3D point cloud — a detailed map of exactly where every surface around the vehicle is, down to fine detail like a pedestrian's outline or a curb's edge. LiDAR's core strength is precise, direct geometric measurement without needing to interpret an image at all. Its core weakness is that laser light scatters off airborne water droplets and ice crystals, so heavy rain, snow, and fog can meaningfully degrade both the range and the reliability of the returned point cloud — exactly the conditions where a driver would want the most help, not the least.

Radar: coarser detail, but it doesn't care about weather

Radar takes a different approach, sending out radio waves instead of light and measuring their reflections. Radio waves at automotive radar frequencies pass through rain, fog, snow, and airborne dust with far less attenuation than laser light does, which makes radar the most weather-resilient sensor in the stack. Radar also has a trick LiDAR and cameras don't offer directly: it measures velocity straight from the reflected signal via the Doppler shift, the same frequency-shift effect that makes a passing ambulance siren change pitch. That means radar can report not just where an object is, but how fast it's moving toward or away from the vehicle, instantly and without comparing multiple frames over time. The trade-off is resolution — radar's reflections are coarser and less detailed than a LiDAR point cloud or a camera image, which makes it much better at answering "is something out there and how fast is it closing" than "exactly what shape is it."

Cameras: the only sensor that reads meaning

Cameras are the odd one out in a useful way: they don't measure distance directly at all, but they're the only sensor in the stack that captures color, texture, and fine visual detail. That's essential for tasks no laser or radio pulse can do — reading a stop sign's text, distinguishing a red traffic light from a green one, recognizing a painted lane marking, or classifying whether a shape ahead is a plastic bag or a small animal. Depth can be estimated from cameras too, either by using stereo pairs or by applying machine-learning models trained to infer distance from a single image, but that estimate is generally less direct and less precise than LiDAR's timed laser pulses. Cameras also share an obvious weakness with human eyes: they're degraded by glare, darkness, dirty or fogged-up lenses, and low-contrast conditions like a white truck against an overcast sky.

Ultrasonic sensors: the close-range specialist

For very short-range tasks like parking, autonomous vehicles typically add a fourth sensor type: ultrasonic sensors, which send out high-frequency sound pulses and time their echoes, much like LiDAR but with sound instead of light. Ultrasonic sensors are inexpensive and reliable at very close range, which is exactly the job they're given — detecting a curb or a parked car's bumper a few feet away — rather than being asked to see a hundred meters down the highway.

Why redundancy, not a single "best" sensor, is the point

Put the three main sensor types side by side and a pattern emerges: LiDAR gives precise geometry but struggles in bad weather, radar shrugs off bad weather but sees coarsely, and cameras read meaning that neither of the others can but fail in glare, darkness, or dirt. No single sensor is strictly better than the others across every condition a real road throws at a vehicle. That's the actual argument for sensor fusion — not that more sensors are always better in the abstract, but that combining overlapping, differently-weak sensor types lets a system cross-check one sensor's reading against another's and keep functioning when any single sensor's conditions turn unfavorable.

See how these systems have evolved

The sensor suites on production and prototype self-driving vehicles have changed considerably as each of these technologies has matured and come down in cost. You can browse the hardware evolution of real self-driving programs, sourced from Wikipedia, in the self-driving wing of the DriveForward Museum.