robotics//sensor//lidar

A lidar is a sensor that measures distances by sending pulses of laser light and timing their return, sweeping them over the scene to build a cloud of 3D points, and it gives warehouse robots, mapping vehicles and autonomous cars the geometry of their surroundings to the centimetre, at night as well as by day. The principle is **time of flight**: a pulse that returns after \(\Delta t\) has travelled to the target and back at the speed of light \(c\), so the range is

A lidar is a sensor that measures distances by sending pulses of laser light and timing their return, sweeping them over the scene to build a cloud of 3D points, and it gives warehouse robots, mapping vehicles and autonomous cars the geometry of their surroundings to the centimetre, at night as well as by day. The principle is time of flight: a pulse that returns after Δt\Delta tΔt has travelled to the target and back at the speed of light ccc, so the range is

r=c Δt2.r=\frac{c\,\Delta t}{2}.r=2cΔt​.

A centimetre of range is about 67 picoseconds of round trip, which is the timing precision the electronics must reach. A rotating lidar sweeps the scene 10 to 20 times a second and produces from hundreds of thousands to millions of points per second; a 2D lidar scans a single plane, enough for a robot moving on a flat floor.

It fails where light is scattered, absorbed or mirrored: rain, fog and dust return false echoes or none, black surfaces absorb the pulse, and glass and mirrors send it elsewhere, so a glass door can be invisible. A radar sees through the weather that blinds it, and a camera reads the colours and signs it cannot (perception).

The cloud is skewed by the vehicle's own motion. A sweep takes 50 to 100 ms, and if the vehicle moves or turns during it, the first and last points are measured from different poses; the scan is corrected (deskewed) with the IMU and the timestamp of each point (time synchronization).

Its points are the raw material of localization. Matching successive scans gives odometry, matching scans against a stored map places the robot with a particle filter, and doing both at once is lidar SLAM, industry in warehouse robots and vacuum cleaners. Like every perception sensor it needs its mounting pose and time offset calibrated (sensor calibration); solid-state lidars without moving parts are a niche advancing quickly.