Vehicles with automated functions require sensors to perceive the environment around them. The sensor suite used usually contains cameras, radar sensors, and LiDAR sensors. Testing automated functions only in the real-world requires a large amount of time and resources.
Therefore, there is a benefit to testing automated functions in simulation as well. Hence, there is a need to replicate things found in the real-world, within simulation. For LiDAR sensors, this means creating a model that replicates the functionality.
LiDAR sensors work by emitting laser beams of infrared light into the environment, which reflects off objects and returns to the LiDAR sensor’s receivers. This received light can be used to understand more about the object that reflected the beam, such as its distance from the LiDAR and its reflectivity. Combining information from hundreds of thousands of emitted beams per scan, builds a “picture” of the surrounding environment called a point cloud.
At Claytex, ray tracing is used to track simulated LiDAR beams through the environment and replicate the characteristics of the light as it returns to the LiDAR sensor. Realistic physics can be integrated to model the reduction in intensity caused by the propagation and reflection, allowing point clouds to be produced that are representative of the real world.
Figure 1 shows a point cloud of an urban environment creating using a LiDAR model within AVelevate.
However, weather effects, such as rain, can change the LiDAR beams propagation through the environment. The LiDAR beams interact with the rain drops, reducing the intensity of light that reaches the receiver. Since raindrops are of various sizes and are not guaranteed to intersect with LiDAR beams as they fall, they present a challenge for modelling of LiDAR sensors.
How do you decide whether a beam interacts with a raindrop?
The size of the raindrop will affect how the light interacts with the drop. How do you decide the size of the raindrop?
At Claytex, we use statistical distributions and physics to model positions and sizes of raindrops and their interaction (or not) with LiDAR beams. Combining accurate physics with statistical distributions allows the modelling of rainfall interaction with LiDAR sensors in a computational efficient and accurate manner, without the need to model the physics of rainfall itself.
Written by Jonathan Robinson, Simulation Engineer
Please contact us to find out more.