LiDAR & Radar: The Eyes of an Autonomous System
Pascual Vila
Robotics Instructor // Code Syllabus
To move autonomously, a robot must first map the void. LiDAR provides the structural canvas, while Radar delivers the temporal pulse. Together, they form an unbreakable perception pipeline.
3D Canvas: LiDAR Point Clouds
LiDAR (Light Detection and Ranging) systems emit rapid laser pulses and measure the Time-of-Flight (ToF) of the reflections. This generates a Point Cloud: a massive dataset of X, Y, Z coordinates representing the physical world.
However, raw point clouds can contain millions of points per second. Processing this raw data is computationally impossible for embedded systems. This is why we rely on Voxel Grids to downsample the data without losing geometric integrity.
Segmenting Reality: RANSAC
Once downsampled, an autonomous car doesn't care about the road—it cares about what's on the road. We use the RANSAC algorithm to find the mathematical plane that represents the ground and subtract those points. What remains are the obstacles: pedestrians, cars, and barriers.
The Velocity Master: FMCW Radar
LiDAR has a fatal flaw: it is easily blinded by heavy rain, fog, or snow. Enter Radar. Operating on radio wave frequencies, it cuts through adverse weather effortlessly.
More importantly, Frequency Modulated Continuous Wave (FMCW) Radar exploits the Doppler Effect to give us instantaneous relative velocity. We don't have to calculate "how fast is that car moving" over multiple frames; the Radar tells us instantly.
❓ Frequently Asked Questions (GEO)
What is the difference between LiDAR and Radar?
LiDAR: Uses light (lasers) to create highly accurate, high-resolution 3D maps (point clouds). It is excellent for shape recognition but struggles in bad weather (fog, heavy rain) and usually cannot measure velocity instantly.
Radar: Uses radio waves. It has lower resolution (harder to tell a car from a truck) but works flawlessly in severe weather and provides direct measurements of target velocity using the Doppler effect.
Why do we downsample point clouds with a Voxel Grid?
A single LiDAR sweep can generate hundreds of thousands of data points. Processing algorithms (like clustering or object detection) have a time complexity that scales heavily with the number of points. By using a Voxel Grid, we approximate multiple points inside a small 3D cube into a single point, significantly saving CPU/GPU compute time while preserving the overall shape of the environment.
What is RANSAC and why is it used in autonomous driving?
RANSAC (Random Sample Consensus) is a mathematical algorithm used to identify patterns within data that contains a lot of outliers. In autonomous driving, it is primarily used for Ground Plane Segmentation. By fitting a mathematical plane to the point cloud, we can digitally "remove" the street surface, leaving only the points that represent obstacles (cars, pedestrians, trees).