How Self-Driving Cars Detect Dirty Lidar Sensors

Waymo's 2024 patent on a self-driving car system that uses two scans from a rotating Lidar sensor to spot dirt or obstructions on its lens.

What it covers

This patent describes a method for a self-driving car's Lidar system to detect when its view is blocked. The Lidar sensor, which uses light to measure distances, has a housing that spins. As it spins, it sends out light pulses and measures how they bounce back, creating a 'scan' of its surroundings. This patent claims a system that takes two separate scans of the same area. By comparing these two scans, the system can identify if a part of the view is consistently missing or distorted, indicating something is blocking the Lidar's optical window, like dirt or a sticker. For example, if the first scan shows a clear road but the second scan shows a blurry patch in the same spot, the system knows something is obstructing the sensor.

What it doesn't cover

• —Lidar systems that do not rotate their sensor housing.
• —Methods that only take a single scan of the environment.
• —Detecting occlusions using sensors other than Lidar.
• —Systems that do not compare multiple scans to identify blockages.
• —Identifying occlusions that are not on the optical window of the Lidar.

The clever bit

Instead of relying on external sensors to check for dirt, the system cleverly uses the Lidar's own scanning process. By comparing two scans taken from slightly different perspectives or at different times, it can infer an obstruction based on inconsistencies in the data itself.

Why it matters

Lidar is crucial for self-driving cars to 'see' their environment. If the Lidar's sensors are dirty or blocked, the car's perception system can be severely impaired, leading to unsafe driving. This patent addresses a fundamental challenge in maintaining the reliability of these critical sensors in real-world conditions.

Real-world examples

• 1.Waymo self-driving vehicles
• 2.Lidar systems on autonomous vehicles