Light pulses, microseconds, and millions of points per second — the sensor that lets a car build a 3D map of everything around it. Tap, drag, and explore.
LiDAR = Light Detection and Ranging （激光雷达 / 光探测与测距）. It fires a laser pulse （激光脉冲）, waits for the reflection （反射）, and measures distance from how long the round trip （往返时间） took.
Light travels at almost exactly 300,000 km/s （光速）. If a pulse comes back in t seconds, the object is (speed × t) ÷ 2 away — divided by two because the light went there and back. Drag the target below and watch the timing change.
Notice how the timer barely moves — even at 250 m the round trip is under 2 microseconds. That's why a LiDAR can fire and listen millions of times per second.
Almost every LiDAR uses one of two physics tricks to turn light into distance.
Sends a sharp pulse and literally times the echo with a stopwatch. Simple, mature, cheap — and what nearly every car on the road today uses.
In production right now, ToF dominates （飞行时间法目前主导市场） — it's mature, affordable, and in volume production. FMCW measures velocity directly via the Doppler shift （多普勒频移，可直接测速度） and resists interference （抗干扰）, but it's pricier and still mostly in sampling （仍处于样品阶段）.
Each returned pulse is one 3D point. Sweep the beam （扫描光束） fast enough and the points form a "point cloud" （点云） — a live 3D scan of the street. Drag to rotate.
Colour here encodes distance — near points warm, far points cool. A real automotive LiDAR generates this for the whole 360° scene up to 10–20 times every second.
Self-driving stacks fuse three sensors. Each covers the others' weaknesses. Toggle them to see what each one "sees."
Precise 3D shape and distance, day or night, no guessing. It maps exactly where a curb, pedestrian, or stopped car is in space — something cameras infer and radar sees only coarsely.
It can't read a red light, a sign, or lane paint — that's the camera's job （摄像头）. And heavy rain, fog, or snow scatter the laser, where radar （毫米波雷达） stays reliable. Fusion （传感器融合） is the answer.
The autonomy ladder （自动驾驶分级） matters here. L2 systems assist a watching driver （辅助驾驶）; L3 lets the car drive itself in limited conditions （有条件自动驾驶）; L4 robotaxis （无人出租车） remove the driver entirely. Higher levels lean harder on LiDAR for a redundant （冗余）, geometry-true view. China's regulators folded LiDAR into mandatory L2 ADAS standards （高级驾驶辅助系统） in 2025, pushing it from luxury option toward standard kit.
They differ in how they aim the beam. Tap each to open its specs.
The classic rooftop "spinning bucket." A motor physically rotates the laser array （激光阵列） for a full 360° view. Superb coverage and range — but bulky, expensive, and moving parts （运动部件） wear out. Dominant on robotaxi test fleets (Waymo-style), fading in passenger cars.
A tiny silicon micro-mirror （微振镜） tilts to steer the beam — no big spinning motor. Much smaller and cheaper than mechanical, a popular middle ground for forward-facing automotive units. The mirror is the only moving part.
Steers the beam electronically with zero moving parts （无运动部件） — far more durable and mass-manufacturable （耐用、易量产）. Now ~58% of the market and the direction the industry is heading. Costs are falling toward ~$500/unit at scale, a 60–70% saving vs. mechanical.
Lights the whole scene at once with one broad pulse — like a camera flash （一次泛光照亮整个场景） — and a detector grid reads all returns simultaneously. No scanning, very fast, great for short-range. Range is limited because energy spreads out; best for close-in and cabin sensing.
A physics category as much as a form factor: instead of pulses, it sends a continuous frequency-swept wave and reads the Doppler shift （多普勒频移）, so it gets velocity per point （逐点测速） instantly and shrugs off sun and other LiDARs' interference. The future-facing bet — still costly and mostly sampling, not volume.
The numbers engineers actually compare. Swipe the table sideways on mobile.
| Type | Field of view | Range | Moving parts | Maturity | Cost band |
|---|---|---|---|---|---|
| Mechanical | 360° | 200–300 m | Yes (motor) | Mature | High |
| MEMS | ~120° | 150–250 m | Tiny mirror | Mature | Moderate |
| Solid-state | ~120° | 150–250 m | None | Scaling | Low → falling |
| Flash | Wide fixed | ~50 m | None | Niche | Moderate |
| FMCW | ~120° | Long | Varies | Emerging | Premium |
← swipe to see all columns →
Range （探测距离） — farthest it reliably detects a low-reflectivity object. FOV （视场角） — how wide an angle it covers. Resolution （分辨率） — points per degree; higher = sees smaller/farther objects. Frame rate （帧率） — full scans per second (10–20 Hz typical). Wavelength （波长） — usually 905 nm (cheap) or 1550 nm (longer range, eye-safe （人眼安全） at higher power).
The headline story of the last two years is collapse — LiDAR went from a luxury add-on to something cheap enough for $15–20k EVs.
Since 2019, Hesai says it cut per-unit cost by about 99.5% to roughly $200. RoboSense's chip-based MX dropped under $200 by replacing FPGAs with custom ASICs （用定制芯片替代通用芯片以降本）. At these prices, LiDAR adoption on Chinese EVs above ~$20k is heading toward 40%, and the automotive LiDAR market is forecast to reach roughly $9.5 billion by 2034.
Tesla famously bets on cameras alone, calling LiDAR a crutch. Most of the rest of the industry — especially in China — bets the opposite: now that a sensor costs ~$200, the geometric certainty is worth it. Both views are live; pricing is what changed the math.