Self-Driving Cars: The LIDAR

That big, bulky component you see on the tops of self-driving cars is called the LIDAR. The LIDAR, which stands for Light Detection and Ranging, is a newer approach to object mapping objects in space. It’s a lot like a radar, but uses light waves to measure distances to objects, rendering much more precise distance data.

The Sensor

The LIDAR allows a self-driving car to view with continuous, full 360 degrees of visibility. A human may forget to look over his shoulder while making a lane change, while the LIDAR provides a constant stream of realtime, accurate, and precise data. The crazy-accurate depth information, in fact, has an accuracy of around ±2cm! At any given moment, the car would know where an object is located with extreme accuracy, and can instantly adjust or compensate for errors made by the car itself or even other drivers.

It works by spinning continuously and firing high-frequency beams of concentrated light, and measures how long it takes for these beams to return to the sensor.

Here’s a good diagram from self-driving car company Voyage:

By creating millions of measurements in fractions of a second, the LIDAR can create intricate 3D visualizations of the world around them in the form of, for example, a “point cloud,” which looks something like the following:

Camera-Only?

A number of companies are trying to push for a self-driving car with no LIDAR, using only cameras and radar sensors. Among these companies is Tesla; Elon Musk brings up a good point in asking the question: If humans don’t need complex and expensive LIDAR sensors to be able to perceive and navigate the world, why can’t an autonomous car?

While companies like Comma.ai and AutoX are achieving impressive results, ultimately the question of safety will outweigh the scalability and low cost. LIDARs cost around $80k, a component more expensive than most regular cars themselves. Sure, humans don’t have LIDARs, but then again, human driving is incredibly accident-prone, with over 3,200 automobile fatalities per day. And, research into lowering LIDAR cost has yielded promising results, with the implementation of multiple solid-state LIDARs as opposed to scanning ones.

An optimal combination of multiple of these under-$1k solid-state LIDARs could cover the full 360 degree range and yield similar results to the more expensive scanning LIDARs.

Not Just Object Detection — Mapping and Planning Ahead

The LIDAR also allows you to generate enormous 3D maps and point clouds, which you can then navigate the car within. By using a LIDAR to map and navigate an environment, an autonomous car knows ahead of time the bounds of a lane, or that there is a traffic light way ahead. This advanced predictability allows for better adjustment and more safety.

The LIDAR is an essential component to the self-driving car, and the advance of LIDAR technology in recent years is what has allowed self-driving cars to operate with such high levels of precision and accuracy.