How Self-Driving Cars See: Lidar, Radar and Cameras

Here is how self-driving cars see, in one sentence: they time the return of laser pulses, bounce millimetre-wave radio off moving metal, and point a few dozen cameras at everything. Waymo’s Jaguar I-PACE carries 29 cameras alone, giving a “simultaneous 360° view around the vehicle.”

Three sensor types, because each one has a failure mode the others do not. That is the whole design, and it is explainable to a nine-year-old with a flashlight and a hallway.

Key Takeaways

• Lidar measures distance by timing a light pulse’s round trip. Waymo describes sending “millions of laser pulses in all directions, then measur[ing] how long it takes for them to bounce back off objects.”
• Radar uses millimetre-wave frequencies to deliver “an object’s distance and speed” and is “effective in rain, fog, and snow”, where light-based sensors struggle.
• Cameras read colour, text and traffic lights, and can spot objects “even from hundreds of meters away,” but need light and cannot measure distance directly.
• Radar’s classic weakness is stationary objects. NHTSA’s 2026 investigation into Comma’s devices found they failed “to detect or respond to slow or stopped vehicles in the same road lane.”
• Light travels about 30 centimetres per nanosecond, which is why lidar is simple in concept and hard in electronics.

Lidar: timing light

Lidar stands for Light Detection and Ranging. NOAA, which uses it to map coastlines, defines it as “a remote sensing method used to examine the surface of the Earth” that “uses light in the form of a pulsed laser to measure ranges (variable distances).” Those pulses “generate precise, three-dimensional information about the shape of the Earth and its surface characteristics.”

The mechanism is a stopwatch. Fire a short laser pulse. It travels out, hits something, scatters, and a tiny fraction comes back. Measure the elapsed time, divide by two, multiply by the speed of light, and you have the distance to that point. Do it millions of times per second across many angles and you have a point cloud, a three-dimensional model of everything around the car.

Now the number that makes this hard. Light covers roughly 30 centimetres per nanosecond. An object 150 metres away sits at a round trip of 300 metres, which takes about one microsecond. To resolve that distance to within 10 centimetres, your timing electronics need to distinguish events less than a nanosecond apart. Conceptually trivial. Electronically brutal.

Waymo says its lidar sensors are “located all around the vehicle,” and that lidar gives the system “a bird’s eye view” regardless of time of day. That last part is the key advantage: lidar brings its own light, so darkness changes nothing.

Lidar’s weaknesses are physical. Rain, fog and snow scatter the beam, since water droplets are excellent at redirecting light. And lidar sees shape, not meaning: a point cloud cannot tell you what colour a traffic light is.

NOAA’s own practice shows how wavelength choice matters: “Topographic lidar typically uses a near-infrared laser to map the land, while bathymetric lidar uses water-penetrating green light to also measure seafloor and riverbed elevations.” Different wavelength, different material penetrated. Same stopwatch.

Radar: seeing speed through weather

Radar sends out radio waves instead of light. Waymo describes its radar as using “millimeter wave frequencies to provide the Waymo Driver with crucial details like an object’s distance and speed,” and notes it “is effective in rain, fog, and snow.”

Two things make radar structurally different from lidar. First, the wavelength is far longer, millimetres rather than hundreds of nanometres, so raindrops are too small to scatter it meaningfully. Radar sees through weather that blinds light-based sensors. Second, radar measures velocity directly, by the Doppler shift in the returned frequency. It is not inferring speed from successive positions; the physics hands it the speed.

And here is radar’s famous failure. The world is full of stationary metal: bridges, overhead signs, guardrails, parked cars, drain covers. If a following system braked for every stationary return, it would be unusable. So these systems weight moving targets and discount stationary ones, which works beautifully until the object ahead is a car that has stopped.

That is not a theory. In September 2026, NHTSA’s Office of Defects Investigation opened case PE26007 into Comma.ai’s aftermarket driver-assistance devices after five crashes, finding they failed “to detect or respond to slow or stopped vehicles in the same road lane.” We cover that case in the federal hands-off driving investigation.

Cameras: the only sensor that reads

A camera measures light intensity and colour across a grid of pixels. That is all. Everything else (this is a stop sign, that is a pedestrian, the light is green) is inference performed by software on top of a flat image.

Which makes cameras uniquely capable in one respect: they are the only sensor that can read. Traffic lights, lane markings, speed limit signs, brake lights, a construction worker’s hand signal. None of that exists in a lidar point cloud or a radar return. Waymo notes its cameras can spot objects “even from hundreds of meters away” and provide a simultaneous 360-degree view across 29 units on the I-PACE.

Cameras fail in the ways eyes fail. They need light. Direct sun into the lens washes out the frame. A tunnel exit swings from dark to bright faster than exposure control can follow. And a single camera cannot measure distance at all, it must infer it from two cameras’ disparity, from motion, or from learned size priors, all of which can be wrong.

The three sensors side by side

	Lidar	Radar	Camera
What it measures directly	Distance (time of flight)	Distance and speed (Doppler)	Light intensity and colour
Brings its own light	Yes	Yes (radio)	No
Works in darkness	Yes	Yes	Poorly
Works in heavy rain/fog/snow	Degraded	Yes	Degraded
Reads text, colour, signals	No	No	Yes
Measures speed without tracking	No	Yes	No
Classic failure	Weather scatter; no semantics	Stationary objects filtered out	Needs light; distance is inferred
Typical cost	Highest	Lowest	Low

Read the “classic failure” row across. Lidar’s failure is weather. Radar’s failure is stopped objects. The camera’s failure is darkness and distance. No two of them fail in the same way, which is the entire argument for carrying all three.

That combination is called sensor fusion, and the point is not redundancy for its own sake. It is that radar covers lidar’s fog problem, lidar covers the camera’s distance problem, and the camera covers radar’s “what even is that” problem.

How to Teach Your Kid About How Self-Driving Cars See

Ages 5–8: the flashlight and the hallway

Two experiments, five minutes each. First, in a dark room, give them a flashlight and ask them to find a specific toy. Then turn on the lights. Point out that one of those was the camera’s problem and the flashlight was lidar’s solution, bring your own light.

Second, go to a hallway, stairwell or parking garage and clap once. Listen for the echo. Explain that the sound went out, hit the wall, and came back, and that if you could time it, you could figure out how far away the wall is. That is the whole idea of lidar with sound instead of light.

Ages 9–12: measure a wall with an echo

Now do the maths. Sound travels about 343 metres per second in room-temperature air. Find a large flat wall across an open space, clap, and have your kid time the echo with a phone stopwatch. It works best at 50 metres or more, where the delay is around 0.3 seconds.

Distance = (speed × time) ÷ 2. Then measure the wall’s distance with a long tape or a map app and compare.

Then the comparison that lands: light is about 874,000 times faster than sound. So to do the same trick with light at the same distance, your stopwatch would need to resolve nanoseconds instead of tenths of a second. The idea is identical; the engineering is not.

Ages 13+: build the spec sheet and defend a choice

Have your teen build the three-sensor table themselves from primary sources: NOAA’s lidar page for the physics, a manufacturer’s sensor page for the configuration. Then ask them to design a sensor suite for a specific job: a delivery robot on a sidewalk, a tractor in a field, a car on a highway at night in freezing rain. Different answers are correct for each, and the reasoning is the deliverable.

The question to ask: “Which sensor would you delete first, and what would the car stop being able to do at 2 a.m. in the rain?”

If they delete the camera, they lose traffic lights. If they delete lidar, they lose precise shape in the dark. If they delete radar, they lose the rain. Every answer costs something specific, and naming the cost is the sign the concept landed.

What to do at home

Point at the hardware on a real vehicle

If a robotaxi operates in your city, look at one. The spinning or fixed lidar units, the camera clusters, the radar behind the bumper fascia. Seeing 29 cameras as physical objects beats any diagram, and kids remember hardware they have stood next to.

Use your phone’s own sensors as the demo

Modern phones with a depth sensor use a version of time-of-flight ranging for face unlock and portrait mode. If yours has one, measure a room with a free lidar-scanner app and then check it with a tape measure. The error you find is the honest version of the lesson.

Connect it to weather you actually have

The next time it fogs or snows, ask which sensor would be struggling right now and which would be fine. Weather turns an abstract table into a lived observation, and it is the fastest route to understanding why fusion exists.

Separate measuring from understanding

This is the deepest idea in the article and worth saying out loud: none of these sensors understands anything. They produce numbers. All meaning is added afterwards by software. A kid who grasps that distinction will not be fooled by “the car sees a pedestrian” language, which hides an entire inference step.

What not to do

Do not teach that more sensors always means safer. Sensor count is not a safety metric. What matters is whether the failure modes are covered and whether the software correctly weighs disagreeing inputs, a harder problem than adding hardware, and one of the places where real engineering effort goes.

What to Watch For Over the Next 3 Months

• Week 4: Watch for any new sensor configuration announcement from an operator, particularly lidar unit counts and whether units are spinning or solid-state. Solid-state lidar with no moving parts is the cost and reliability frontier for this hardware.
• Month 2 red flags: Watch for claims that one sensor type alone is sufficient. Those claims are an engineering position, not a settled fact, and the honest framing names which failure mode is being accepted in exchange.
• Month 3 self-check: Can your kid explain, without the table, why a driverless car carries three different sensor types? If the answer is “to be safe,” push once more. The answer you want is about different failure modes, and it is a better answer than most adults give.

Frequently Asked Questions

What is lidar in one sentence a kid can repeat?

A laser stopwatch: it fires light pulses, times how long they take to bounce back, and turns those times into distances. NOAA defines it as using “light in the form of a pulsed laser to measure ranges.”

Why not just use cameras, since humans only have eyes?

Humans also have forty years of world knowledge, two eyes with stereo depth, a neck that moves, and the ability to slow down when confused. A camera-only system has to infer distance from a flat image. It can be done, and it is an active engineering position, but it means accepting the camera’s failure modes, darkness, glare, inferred distance, without a sensor that covers them.

Does rain break lidar?

It degrades it. Water droplets scatter light, so returns get noisy. That is precisely why radar is in the suite: Waymo describes its radar as “effective in rain, fog, and snow,” covering the condition where lidar is weakest.

Why does radar miss stopped cars?

Because stationary radar returns are mostly clutter, signs, bridges, guardrails, parked cars, so following systems discount them. NHTSA’s PE26007 investigation found devices failing “to detect or respond to slow or stopped vehicles in the same road lane,” which is this filtering problem showing up in the real world.

How many sensors does a real robotaxi have?

Waymo publicly states 29 cameras on its Jaguar I-PACE vehicles, plus lidar units “located all around the vehicle” and millimetre-wave radar. Exact counts vary by hardware generation and company.

Is any of this relevant if we will never own a driverless car?

Yes, because the same sensors are in phones, vacuum robots, drones, warehouse equipment and increasingly in ordinary cars’ driver assistance. Understanding time-of-flight ranging and the difference between measuring and understanding transfers across all of it, as does knowing which automation level your own car is.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. National Oceanic and Atmospheric Administration. “What is lidar?” NOAA Ocean Service. https://oceanservice.noaa.gov/facts/lidar.html
2. Waymo. “Waymo Driver.” https://waymo.com/waymo-driver/
3. TechCrunch. (2026, September 23). “Comma’s hands-off driving tech under investigation after 2 fatal crashes.” https://techcrunch.com/2026/09/23/commas-hands-off-driving-tech-under-investigation-after-2-fatal-crashes/
4. National Highway Traffic Safety Administration. “Driver Assistance Technologies.” https://www.nhtsa.gov/vehicle-safety/driver-assistance-technologies
5. Insurance Institute for Highway Safety. “Advanced driver assistance.” https://www.iihs.org/topics/advanced-driver-assistance
6. NGSS Lead States. “Appendix I — Engineering Design in the NGSS.” https://www.nextgenscience.org/resources/ngss-appendices
7. Insurance Institute for Highway Safety. (2026, July 23). “How safe are driverless cars?” https://www.iihs.org/news/detail/waymos-driverless-cars-crash-less-often-than-people