Repository navigation
Inquiry About Realistically Simulating LiDAR Performance Degradation in Adverse Weather Conditions in CARLA #10123
Replies: 2 comments
|
Good project idea, and you are right to be suspicious of the weather settings. Short version for each question: 1. What CARLA's sensors modelThe weather parameters ( What the LiDAR does give you are knobs you can set yourself:
Raising The radar has no weather model. In reality radar is affected much less than LiDAR by fog and only moderately by heavy rain, so "radar unchanged" is not a bad approximation. That asymmetry is exactly why your LiDAR vs radar vs fusion comparison is interesting. 2. Validating against real dataReal datasets that have LiDAR in bad weather:
Don't try to match absolute numbers. Real sensors and CARLA's sensor model are too different for that. Compare the relative change from clear to fog/rain, using the same metrics in both:
If the curve "share of points kept vs distance" for your simulation has the same shape as the real one at a similar visibility, you have a reasonable argument for your method. 3. Doing it realistically in PythonThe most common approach in research is: record clear-weather point clouds in CARLA, then apply a physics-based post-processing step in Python. These have open code:
A simple first version you can write yourself is to link fog to visibility. Meteorological visibility V and the extinction coefficient α are related by α ≈ 3 / V (Koschmieder). The returned power goes out and back, so it scales with exp(-2αd). For each point, multiply the intensity by exp(-2αd), drop the point if it falls below a threshold, and optionally add a few random near-range points for backscatter. Then sweep V (e.g. 1000, 200, 100, 50 m) and plot your metrics. This is easy to explain in a report, and you can later compare it against the LiDAR_fog_sim results. A setup that should work well for a school project:
Step 4 is where your "simulation vs reality" goal comes in, and it will make the project stand out. |
|
To add the physics side to the reply above: what fog and rain do to a LiDAR pulse is mainly two-way extinction plus backscatter from droplets.
|
Uh oh!
There was an error while loading. Please reload this page.
Hello,
I am a second-year high school student from South Korea, interested in autonomous vehicle safety and sensor performance.
I am currently planning a research project using CARLA and Python to analyze how LiDAR performance changes under different weather conditions, including clear weather, fog, and rain. I also intend to compare the results of LiDAR-only detection, Radar-only detection, and LiDAR-Radar sensor fusion.
However, I have learned that changing the weather settings in CARLA may not fully reproduce the physical effects of fog and rain on LiDAR measurements, such as signal attenuation and noise. This has led me to wonder whether simply changing the weather settings is sufficient to accurately analyze sensor performance under real-world adverse weather conditions.
I would greatly appreciate your advice on the following questions:
Through this project, I hope to go beyond simply comparing simulation results. My goal is to understand the differences between simulated and real-world conditions and investigate what factors need to be considered to make the simulation more realistic.
Thank you very much for taking the time to read my inquiry. Any guidance or resources you could provide would be extremely helpful for my research.
Best regards,
A High School Student from South Korea
All reactions