ApolloSim: A Lidar Simulator With Calibrated Sensor Noise

Abstract

In this work, I present a light detection and ranging (lidar) sensor simulator that incorporates realistic sensor noise. Lidar sensors are used for various applications, ranging from autonomous vehicles to satellite imagery. Field testing a lidar presents multiple inconveniences and challenges; it is often costly, poses a risk of damage to the device, and is time-intensive. Therefore, the benefits of employing a lidar simulator are evident. Moreover, lidar data can be crucial for training large-scale deep learning models, which require massive datasets; generating this data in simulation is significantly quicker and more cost-effective than manual data collection.

I introduce a lidar simulation methodology that uses a real lidar sensor to calibrate the simulation, in order to maximize the accuracy of the simulation. The simulation utilizes sensor data from multiple benchmark materials to accurately replicate the noise in the data produced by the sensor. The results indicate that simulations calibrated with real sensor noise outperform those based on standard parametric approaches. The data generated by the simulation confirms that calibrating a lidar simulation with the target sensor is a viable and economical approach for rapid and precise lidar simulations.