LiDAR

LiDAR, which stands for Light Detection and Ranging, is a way of measuring distance with light. The sensor emits short pulses of laser light and measures the time it takes for each pulse to bounce off a surface and return, and since the speed of light is known, that time directly gives the distance to the surface. By sending out many pulses in many directions and sweeping across a scene, LiDAR builds up a dense collection of measured 3D points, a point cloud, that traces the shapes of everything around it with high geometric accuracy.

This direct, precise measurement of 3D structure is what makes LiDAR so valuable for physical AI. Where a camera captures rich color and texture but has to infer depth, LiDAR measures depth outright, producing a faithful geometric picture of the environment. That is why it is a primary sensor on many autonomous vehicles and robots, typically used alongside cameras and radar so their strengths complement each other. Its main limitations are that it can be degraded by heavy rain, fog, or dust, and it produces large volumes of data, but its accuracy in capturing 3D shape is hard to match.

A LiDAR sensor fires laser pulses and measures the time of flight for each reflection, converting that timing into a distance for the point the pulse struck. By scanning the beam across the field of view, whether by rotating, using mirrors, or other means, it samples many directions and assembles the returns into a point cloud, where each point has a 3D position and often an intensity value describing how strongly the pulse reflected. The measurements are expressed in the sensor's own frame, so calibration is needed to relate them to other sensors and to a common coordinate system. The result is a spatially accurate but sparse and unevenly distributed representation of the scene.

LiDAR matters because accurate 3D geometry is essential for machines that must navigate and act in physical space, and LiDAR provides that geometry directly rather than by inference. For anyone working on autonomous vehicles or robots, it is a cornerstone of perception, and much of 3D annotation and detection, such as fitting cuboids to objects, is built on the point clouds it produces. Understanding LiDAR, including both its accuracy and its limits in bad weather, is key to understanding how physical AI systems sense depth and structure.

LiDAR uses laser light to produce dense, high-resolution 3D point clouds with accurate geometry, while radar uses radio waves to measure range and velocity more coarsely but with greater robustness in rain, fog, and dust. They are complementary, which is why many systems use both.

Because it measures distance to many individual points by scanning laser pulses across a scene, and each measurement becomes a 3D point. Together these points form a cloud that traces the surfaces around the sensor.

Its performance can degrade in heavy rain, fog, or dust, which scatter the laser light, and it generates large volumes of data. It also provides limited color or texture information compared with cameras, which is one reason it is usually combined with other sensors.

Point cloud, Radar, Multimodal sensor data, Sensor fusion, 3D bounding box / cuboid.