How LiDAR Actually Works: The Physics Explained

Point a flashlight at a wall and you learn nothing about distance. Point a laser that can time its own echo to a billionth of a second, and you can measure that wall to within a few millimeters from across a parking lot. That is the whole trick, and understanding how LiDAR works comes down to one stubborn fact of nature: light travels at a fixed, knowable speed, so the time a pulse takes to fly out and bounce back is a ruler. LiDAR — Light Detection and Ranging — is a laser tape measure that fires hundreds of thousands of times per second and stitches every return into a three-dimensional map of the world. The physics is elegant; the engineering to make it small, cheap, eye-safe, and immune to sunlight is where the real fight lives.

What this covers: the core distance trick, time-of-flight versus FMCW sensing, the laser-and-detector chain, how a point cloud is assembled, beam steering and solid-state designs, the range equation and its hard limits, and where LiDAR beats — and loses to — radar and cameras.

Context: what LiDAR is and where it shows up

A LiDAR sensor measures distance with light. It sends out laser pulses (or a continuous laser beam), waits for the reflection off whatever is out there, and converts the round-trip into range. Sweep that measurement across thousands of directions and you get a point cloud — a dense scatter of XYZ coordinates that traces the literal shape of the scene: the curb, the cyclist, the pallet rack, the contour of a hillside.

The idea is old. Engineers bounced laser pulses off the Moon in the 1960s using retroreflectors left by Apollo, nailing the Earth-Moon distance to centimeters. Aircraft-mounted LiDAR has mapped terrain and forest canopy for decades, and archaeologists have used it to find lost cities under jungle by stripping the trees out of the data. What changed recently is scale and cost. The same physics now lives in self-driving car roofs, warehouse robots, drones, smartphones (the iPhone’s face-mapping and room-scanning sensors are LiDAR), and factory inspection rigs that catch a misplaced weld.

LiDAR earns its keep wherever a machine needs an accurate, direct, three-dimensional measurement of geometry rather than a guess. A camera sees a flat picture and has to infer depth; LiDAR measures it. That single difference is why it became central to robotics and autonomy — and why the rest of this article is about the physics that makes the measurement trustworthy.

The core physics: time-of-flight vs FMCW

Every LiDAR answers the same question — how far? — but there are two fundamentally different ways to ask it, and they shape everything downstream. The diagram below contrasts the two signal chains.

Direct time-of-flight times a sharp pulse and divides by two for range; FMCW sweeps the laser frequency and reads range and velocity from the beat tone where the echo mixes with the outgoing light.

Direct time-of-flight: timing a pulse

The most intuitive approach is direct time-of-flight (dToF). Fire a very short, very bright laser pulse — a few nanoseconds long — start a high-speed clock, and stop the clock when the detector sees the reflection come back. The distance follows from the simplest equation in the field:

range = (c × t) / 2

where c is the speed of light (about 3 × 10⁸ meters per second) and t is the measured round-trip time. The division by two accounts for the fact that the light travels to the target and back, so it covers twice the distance you care about.

The numbers are humbling. Light covers about 30 centimeters every nanosecond. To resolve range to one centimeter, your timing electronics must resolve roughly 67 picoseconds — 67 trillionths of a second. This is why LiDAR is fundamentally a timing problem dressed up as an optics problem. The laser is easy; the stopwatch is hard. Engineers solve it with time-to-digital converters and clever statistics, often firing the same direction many times and histogramming the returns to pull a confident peak out of noise.

dToF is the workhorse of long-range automotive and survey LiDAR. It scales to hundreds of meters because a single high-energy pulse can carry enough photons to survive the round trip and still register against background light.

Indirect time-of-flight: reading a phase shift

A cheaper cousin, indirect ToF (iToF), doesn’t time a discrete pulse. It modulates a continuous light source so its brightness oscillates like a sine wave, then measures the phase shift between the outgoing wave and the returning one. Since the wave’s period corresponds to a known travel distance, the phase delay maps to range. This is the technique inside many depth cameras and short-range consumer sensors. It’s elegant and integrates well onto a chip, but it suffers from phase ambiguity — once the target is far enough that the returning wave has slipped a full cycle, the sensor can’t tell a near object from a distant one without extra tricks. That confines iToF to shorter ranges, typically a few meters to low tens of meters.

FMCW: borrowing a trick from radar

The most interesting newcomer is FMCW — Frequency-Modulated Continuous Wave. Instead of timing a pulse, FMCW LiDAR continuously emits a laser beam whose optical frequency is swept up and down in a sawtooth or triangle pattern, a “chirp.” A small fraction of the outgoing light is split off and kept as a reference. When the echo returns, it’s mixed with that reference on the detector.

Here is the beautiful part. Because the laser frequency has been climbing the whole time the echo was in flight, the returning light was emitted earlier and therefore carries a slightly lower frequency than the light leaving right now. Mixing the two produces a low-frequency beat tone — the difference between them — that an ordinary electronic circuit can measure. The beat frequency is directly proportional to range:

range ∝ beat frequency / chirp slope

No picosecond stopwatch required; you’ve converted a brutal timing problem into a gentle frequency-measurement problem. That is FMCW’s first superpower.

The second is velocity. If the target is moving, the Doppler effect shifts the echo’s frequency on top of the chirp shift. By comparing the up-sweep and down-sweep beat tones, FMCW separates the range component from the Doppler component and reports both distance and instantaneous radial velocity in a single measurement — per point, per shot. dToF can only infer velocity by comparing successive frames. FMCW measures it directly, the way police radar measures your speed.

FMCW pays for this with complexity: it needs an extremely pure, stable laser (a wandering frequency corrupts the beat tone) and coherent detection, where the light’s wave phase is preserved. That coherence buys a third advantage — FMCW is naturally immune to sunlight and to other LiDARs’ pulses, because only light that beats coherently against its own reference produces a signal. Random ambient photons just don’t mix in. The trade-off is range performance and laser cost, which is why dToF still dominates while FMCW climbs.

The laser-and-detector chain

Whether the architecture is pulsed or coherent, the same physical chain has to work: generate light, send it out, collect what bounces back, and turn faint photons into a clean electrical signal. The figure below walks the photons from laser to point.

Light leaves the laser, scatters off the target, a tiny fraction is collected by the receive optics, the photodetector converts it to current, and the timing or beat-frequency electronics compute one point.

Choosing a wavelength: 905 nm vs 1550 nm

LiDAR doesn’t use visible light. The two dominant choices both sit in the near-infrared, and the choice between them is a deep engineering fork.

905 nanometers is just past the red edge of human vision. The lasers are cheap silicon-compatible diodes, and crucially, ordinary silicon photodetectors are sensitive at this wavelength — so the whole sensor can be built from low-cost, mature semiconductor parts. The catch is eye safety. At 905 nm the eye’s lens still focuses the light onto the retina, so the power you’re allowed to emit is capped to avoid eye damage. That power ceiling limits range.

1550 nanometers is deeper infrared, in the band fiber-optic telecom built itself around. Its killer feature is biological: water in the front of the eye absorbs 1550 nm light before it can reach and focus on the retina. That means you can legally emit far more power at 1550 nm and stay eye-safe — often a factor of ten or more — which translates directly into longer range and better performance through rain and haze. The price is that silicon is blind at 1550 nm, so detectors must be made from exotic materials like indium gallium arsenide (InGaAs), and the lasers are pricier telecom-derived parts. Most long-range and FMCW automotive systems lean toward 1550 nm; cost-driven and shorter-range systems favor 905 nm.

Detectors: APD, SPAD, and single-photon counting

A returning pulse from 200 meters away may deliver only a handful of photons. The detector’s job is to hear that whisper.

An avalanche photodiode (APD) is a photodiode run at high reverse voltage so that each absorbed photon kicks loose an electron, which then collides its way to an avalanche of electrons — built-in amplification, like a microphone with a gain knob. APDs give an analog signal proportional to the incoming light and are the classic choice for measuring pulse shape and intensity.

A single-photon avalanche diode (SPAD) is an APD pushed past its breakdown voltage, into “Geiger mode.” Now a single photon triggers a self-sustaining avalanche — a full digital click. SPADs are exquisitely sensitive, count individual photons, and pair naturally with histogramming: fire thousands of pulses, log the arrival time of every photon click, and the true target return piles up as a sharp peak above the flat carpet of random sunlight clicks. Arrays of SPADs built on standard CMOS are a big reason flash and solid-state LiDAR became feasible — you can put thousands of single-photon timers on one chip. FMCW systems instead use coherent (balanced) detectors that listen for the beat tone rather than counting raw photons.

Building a point cloud and steering the beam

One measurement gives you one number: the distance straight ahead. To build the 3D point cloud — the thing that actually looks like the world — you have to aim the beam in many directions and record range for each. That aiming problem is beam steering, and it’s the single biggest divider between LiDAR designs.

Beam steering splits into moving-mirror approaches and no-moving-parts solid-state approaches, each trading cost, field of view, range, and durability differently.

From ranges to XYZ

Each shot returns a distance, and the system knows the exact direction it fired in — the azimuth (left-right angle) and elevation (up-down angle) of the beam at that instant. Distance plus two angles is a point in spherical coordinates, trivially converted to an XYZ Cartesian point. Do this hundreds of thousands of times per spin and you get a frame: a snapshot point cloud, typically refreshed 10 to 20 times a second. Stack frames over time, fuse them with the vehicle’s own motion, and downstream software clusters the points into objects, fits the ground plane, and tracks how things move. The raw physics ends at “here is a point”; perception begins there.

Mechanical spinning

The original automotive LiDAR — the spinning bucket on early self-driving prototypes — physically rotates a stack of laser-detector pairs a full 360 degrees, dozens of times per second. It works beautifully: full surround view, long range, mature. But it has motors and bearings that wear, it’s bulky, and at volume it’s expensive. It’s the benchmark everyone is trying to replace.

MEMS mirrors

A microelectromechanical mirror — a tiny silicon mirror, often a millimeter or two across, tilted by electrostatic or magnetic actuation — sweeps a single laser beam across the scene like a raster scan. It’s the half-step to solid-state: there’s still a moving part, but it’s microscopic, robust, and chip-scale. MEMS LiDAR is compact and cheaper than spinning units, at the cost of a narrower field of view per module, so several are tiled to cover the surroundings.

Optical phased arrays: steering with no moving parts

The most physics-rich approach is the optical phased array (OPA), the optical twin of the electronically steered radar antennas on warships. Split the laser across many tiny emitters on a chip and control the phase of each one. By the wave nature of light, when the emitted wavelets interfere they reinforce in one specific direction set by the phase pattern. Change the phases — purely electronically, no motion — and the combined beam swings to a new angle. An OPA steers a beam by rewriting interference, at electronic speeds, with literally nothing moving. It’s the holy grail for durability and integration; the challenge is fabricating arrays precise enough to form a tight beam and suppress stray side-lobes.

Flash LiDAR

Flash LiDAR throws away scanning entirely. It floods the whole scene with one wide laser pulse, like a camera flash, and reads the returns on a 2D array of detectors — each pixel times its own echo. One shot, full frame, no steering. It’s mechanically dead-simple and immune to motion artifacts, but spreading the energy over the entire field means each point gets fewer photons, which limits range. Flash excels at short-range, wide-angle jobs and struggles at the long distances where automotive needs it most.

Limits: the range equation, eye safety, and weather

LiDAR is physics, and physics sets hard ceilings. Three of them define what any sensor can and cannot do.

The range equation: why far is so hard

The number of photons that make it back to the detector falls off brutally with distance. The outgoing beam spreads, only a small patch of the target is illuminated, the target scatters light in all directions, and the receiver lens catches just a sliver of that scatter from far away. Combined, the received power drops roughly as 1 / range² for a broad surface — and even faster for small or angled targets. Double the distance and you may get a quarter of the light, or less.

This is why range is the central battleground. To see twice as far you need far more than twice the laser power, and you’re fighting eye-safety limits the whole way. It’s also why dark, non-reflective objects — a person in black clothing, a matte tire on the road — are the genuinely hard targets: they send back a fraction of the photons a bright road sign would, exactly when you most need to detect them.

Eye safety: the invisible budget

Because LiDAR uses invisible infrared lasers that people stare into without flinching, eye safety is a legal and ethical hard limit, governed by international laser-safety standards. Sensors are designed to stay within “Class 1” — safe under all normal conditions. As covered above, this is the deep reason 1550 nm is attractive: the eye’s own water shields the retina, so the eye-safe power budget is far larger, and more power means more range. Every range spec you read is implicitly capped by how many photons the engineers were allowed to emit.

Sunlight, rain, and fog

The sun is a flood of broadband infrared, and it competes directly with the laser’s faint return — outdoor LiDAR is always working against a bright background. Designers fight back with narrow optical bandpass filters that pass only the laser’s wavelength, very short detection windows, and the histogramming tricks that let SPADs dig a real peak out of solar noise. FMCW sidesteps the problem almost entirely, since incoherent sunlight can’t produce a beat tone.

Weather is harsher. Rain and fog are clouds of droplets that scatter and absorb the beam, sapping range and throwing spurious returns from the droplets themselves. Fog is the worst case: it can both attenuate the signal and create a wall of false near-returns. Longer wavelengths and clever return-filtering help, but no LiDAR sees through dense fog the way radar does — a fundamental reason autonomous systems fuse multiple sensor types rather than betting on one.

LiDAR vs radar vs cameras

No single sensor wins. Each measures the world through a different slice of the electromagnetic spectrum, and their strengths and blind spots are almost perfectly complementary — which is why serious robots and vehicles carry all three.

Cameras are passive, cheap, and rich: they see color, read text and signs, classify objects, and deliver the highest angular detail of any sensor. But they don’t measure depth — they infer it — and they’re at the mercy of lighting, blinded by glare and darkness, and easily fooled by a flat poster of a road. Cameras answer what is it? better than anything else.

Radar uses radio waves, with wavelengths millimeters long instead of LiDAR’s near-infrared. That long wavelength is its gift: radio sails through rain, fog, dust, and snow that stop light cold, and like FMCW LiDAR it reads velocity directly via Doppler. The price is resolution — radar’s long wavelength can’t form the fine, sharp beam light can, so its picture of the world is coarse and blobby. Radar answers is something there and how fast is it moving? in any weather.

LiDAR sits in between: far sharper than radar, and unlike a camera it directly measures precise 3D geometry, day or night, without inferring anything. Its weaknesses are cost, degradation in heavy fog, and trouble with very dark or mirror-like surfaces. LiDAR answers exactly what shape is the world and precisely how far?

The honest engineering conclusion is fusion. Cameras for semantics, radar for all-weather velocity and presence, LiDAR for precise geometry — combined into one model of the world that’s more reliable than any sensor alone. If you’re piecing together the bigger picture, see how these feeds slot into an autonomous vehicle reference architecture, and how a point cloud drives motion planning in a ROS 2 and Nav2 autonomous mobile robot on a warehouse floor.

FAQ

How accurate is LiDAR?
Most automotive and survey LiDAR resolves range to roughly one to a few centimeters, with the best systems reaching millimeter-class precision at short range. Accuracy depends on timing electronics, how reflective the target is, range, and how many shots are averaged. The fundamental limit comes from how finely the system can measure round-trip time — about 67 picoseconds per centimeter of range.

What’s the difference between LiDAR and radar?
Both measure distance with reflected waves, but LiDAR uses near-infrared light while radar uses radio waves millimeters long. The short wavelength gives LiDAR far sharper resolution and crisp 3D geometry; radar’s long wavelength gives it the ability to punch through rain, fog, and dust that LiDAR struggles with. They’re complementary, which is why many systems use both.

Why does LiDAR use infrared instead of visible light?
Infrared is invisible, so it won’t distract drivers or pedestrians, and the wavelengths chosen (905 nm and 1550 nm) align with cheap semiconductor lasers and detectors. The 1550 nm band is especially valued because the eye’s own fluid absorbs it before it reaches the retina, allowing higher, still-eye-safe power and therefore longer range.

Can LiDAR work in rain, fog, or snow?
It works in light rain and snow with some range loss and extra noise, but heavy fog is its weakness — dense droplets scatter the beam and throw false returns. Radar handles those conditions far better, which is the main reason autonomous systems fuse LiDAR, radar, and cameras rather than relying on any one sensor.

What is a point cloud?
A point cloud is the collection of 3D points a LiDAR produces, each with an XYZ position computed from one beam’s measured distance plus the known firing direction. Hundreds of thousands of points per frame trace the literal shape of the scene, which software then clusters into objects, ground, and obstacles.

Is solid-state LiDAR better than spinning LiDAR?
Solid-state designs — MEMS, optical phased arrays, and flash — promise lower cost, smaller size, and no wearing parts, which matters for mass-market vehicles. Spinning LiDAR still leads in range and full 360-degree coverage. The trend is toward solid-state as the technology matures, but the right choice depends on the field of view, range, and budget a given application needs.

Further reading

• How quantum dots actually work: the physics — another first-principles tour of light-and-matter engineering.
• Autonomous vehicle reference architecture — where LiDAR, radar, and cameras fuse into a perception stack.
• ROS 2 and Nav2 autonomous mobile robot navigation — turning point clouds into robot motion.
• NASA: Lunar Laser Ranging — the original time-of-flight measurement, Earth to Moon.
• IEEE Spectrum: LiDAR coverage — ongoing reporting on dToF, FMCW, and solid-state developments.