By Mark Crawford
Lidar, or light detection and ranging, is a popular remote sensing method used for measuring the exact distance to an object. These imaging systems are increasingly used in the development of autonomous vehicles (AVs). However, final technology implementation cannot yet be achieved because major AV manufacturers are only now starting to select providers for data collection units that can be introduced in commercial vehicles.
One of the biggest unknowns in the development of AVs is how adverse weather conditions challenge the development of safe and reliable advanced autonomous systems for sensing and perception. How does the diffraction of light through rain, fog, and snow limit lidar performance?
Although many AV developers are hopeful that automotive lidar sensors will be safe and effective in all weather conditions, a quantitative analysis of these effects is lacking. Currently, testing for autonomous vehicles is primarily performed in sunny environments. Testing conducted in good weather cannot be assumed to correlate directly with performance quality under extreme weather conditions, which could result in false detection caused by the backscattered intensity, thereby reducing the reliability of the sensor. Thus, the sensitivity of lidar to weather is a major drawback to commercialization.
To help resolve this issue, researchers at Frostburg State University in Frostburg, Md., led by Jamil Abdo, associate professor of mechanical engineering and department chair, tested lidar sensors in adverse weather to understand how extreme weather impacts lidar performance.
The results were presented in “Effective Range Assessment of Lidar Imaging Systems for Autonomous Vehicles Under Adverse Weather Conditions with Stationary Vehicles” in ASME’s Journal of Risk and Uncertainty in Engineering Systems.
Approach and Methodology
Abdo and his team designed an experiment to evaluate lidar performance under adverse weather conditions. Data sets were recorded under these conditions to determine the effective range to estimate the maximum range at which the lidar sensors can return points.
The testing set-up and algorithms were developed for two commercial lidar sensors—Velodyne VLP-32 and Ouster OS1-32. Weather resistance of the set-up was an important consideration. The weatherproof platform was designed so that the sensors units were located about three feet high off the ground.
We wrote programs to interface with these lidar units and to analyze the data,” Abdo said. “These programs allowed us to count the number of points in a region of space.”
The team selected a remote area for testing to ensure that other cars did not drive by, affecting the data. The test was also designed to meet the goals of ease of testing and reproducibility in testing locations.
“Even though the western Maryland area around Frostburg is a great location for testing, it was very hard to have a reproducible weather condition,” Abdo said. “We had to use design-of-experiment techniques to account for nondesirable factors such as noise during the testing.”
Abdo decided the best way to test the lidar units was in a worst-case scenario with two cars heading directly toward each other, recreated by driving a car straight toward the test platform. “This is the worst-case scenario for car detection as when the car faces the lidar sensor, its cross-section is the smallest with respect to the sensor,” Abdo said.
Two types of tests were performed under various weather conditions. The first involved stationary tests where the car stopped every 7.5 meters for a few seconds before driving forward another 7.5 meters. The second method conducted testing with the car moving toward the lidar units at varying speeds of five mph, 10 mph, and 20 mph.
Test results showed that both dense fog and heavy rain severely impacted lidar sensors’ object detection and operating range detection. Light rain had minimal effect on lidar performance, and snow had no negative impact.
Moving Forward
Determining the influence of weather on lidar sensor performance is a critical step toward improving the safety levels for autonomous driving under adverse weather conditions “because such analyses provide reliable information for adapting vehicle behavior,” Abdo said.
The AV industry has already taken notice of Abdo’s work. Three major lidar manufacturers have contacted Abdo since he published the team’s research/findings in September 2021, he said. The team has also received funds from Maryland Industrial Partners (MIPS) and Intelligent Fusion Company to support the second phase of research activities.
For this work, “Lidar performance in rain, snow, fog, smog, and dust will be investigated using dynamic sensors and moving targets,” Abdo stated. “We also plan to explore object tracking—for example, other vehicles and any potential obstacles—with lidar systems.”
Mark Crawford writes about technology and engineering from Cerritos, N.M