Lidar vs cameras for self-driving cars: which is better?

Search for "Rivian R2 lidar" or "why doesn't Tesla use lidar" and you land in one of the longest arguments in the car industry: can a car drive itself with cameras alone, or does it need lidar? Here's a plain-English guide to how the sensors work, what each is good and bad at, and which carmakers use what.

The three main sensors

Most driver assistance and self-driving systems are built from some mix of three sensor types.

Cameras capture images much like your eyes. They are cheap, high resolution and the only sensor that reads colour, so they handle traffic lights, road signs, brake lights and lane markings. Their weakness is that a single image has no direct distance information: depth has to be estimated by software, from multiple cameras, motion over time or AI models trained on huge amounts of driving video. Cameras also suffer from glare, darkness, dirt and heavy rain.

Lidar (light detection and ranging) fires short laser pulses and times how long they take to bounce back. Repeated millions of times per second, that builds a precise 3D point cloud of everything around the car. Lidar measures distance directly, works the same at night as in daytime and is very good at the exact shape and position of objects. It cannot read colour, so it can't tell a red light from a green one, and dense fog, heavy snow or spray can scatter the beam. It has also been the most expensive sensor.

Radar sends out radio waves. It measures distance and, thanks to the Doppler effect, the speed of objects directly, and it keeps working in fog, rain and darkness. Traditional automotive radar has low resolution, which makes it poor at telling what an object is. Newer "4D imaging" radars improve this.

Many cars also use short-range ultrasonic sensors for parking, but these play little part in driving at speed.

The camera-only approach: Tesla

Tesla is the best-known camera-only company. It stopped fitting radar to Model 3 and Model Y in North America in 2021, calling the new setup Tesla Vision, and began removing ultrasonic parking sensors in late 2022. Its argument is that roads are designed for human eyes, so cameras plus powerful neural networks should be enough, and that one sensor type avoids conflicts between sensors and keeps costs down across millions of cars.

The open question is the last few per cent of rare, tricky situations: low sun, unusual obstacles, poor visibility. Tesla's FSD (Supervised) is still a Level 2 system that requires an attentive driver, as we explain in our guide to self-driving levels, so the driver is the backup. Crashes such as the FSD case where a driver fell asleep show why that supervision matters.

The multi-sensor approach: Waymo, Mercedes, Rivian and China

Waymo, which runs driverless robotaxis, combines lidar, cameras and radar; its sixth-generation system uses 13 cameras, 4 lidars and 6 radars. With no human backup, it wants redundancy: if one sensor is blinded, another sees the obstacle in a different way.

Mercedes-Benz uses lidar in Drive Pilot, the Level 3 system that lets drivers look away in approved conditions, where the carmaker takes responsibility while the system is active. Volvo launched the EX90 with a roof-mounted lidar.

Rivian has said its next-generation autonomy platform for the R2 pairs a new in-house computer with a front-mounted long-range lidar, plus radar and cameras. Rivian says lidar-equipped R2s go to employees first in late 2026, with customers in 2027, and that eyes-off driving will need the lidar hardware. Rivian's current owners can already try Point to Point driving in early access on existing hardware.

In China, lidar has become common even on affordable EVs, such as the Leapmotor A10S.

Why lidar got cheaper

Early spinning lidars cost tens of thousands of dollars, which is why only research cars and robotaxis carried them. Solid-state and semi-solid-state designs, mass production and Chinese suppliers have pushed prices down sharply. Rivian told investors in 2026 that lidar now costs it "a few hundred dollars", compared with thousands or even tens of thousands a few years ago. Cost is no longer a strong argument against lidar; design, integration and software still are.

Does more sensors mean safer?

Not automatically. More sensors give more data and an independent cross-check, but software has to fuse them correctly, and every sensor needs cleaning and calibration. A badly tuned multi-sensor car can still make mistakes, and a strong camera system can do well in good conditions. What matters for you is:

1. What level the system is. Level 2 means you supervise, whatever sensors it has.
2. Where it's approved. Highways only, mapped cities, certain speeds or weather.
3. Who is responsible. With Level 2 it's you; with approved Level 3 or 4 it can be the manufacturer or operator.