Generating synthetic data for autonomous vehicle applications with the MSU autonomous vehicle simulator

Engineering notes

Key topics: autonomous driving, autonomous vehicle, object detection, lidar, sensor fusion. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

The MSU Autonomous Vehicle Simulator (MAVS) is a versatile and customizable software library for simulating autonomous and unmanned ground vehicles in off-road environments. This article will highlight MAVS’ application to generating synthetic sensor data - including camera and lidar sensors - using physics-based raytracing. In the five years since its release, MAVS has been used to generate synthetic data for many machine learning applications related to autonomous driving including: evaluation of the effects of image compression on image segmentation, object detection and semantic image segmentation in rain, lidar segmentation using CNN in offroad environments, detection of negative obstacles using camera and lidar data, multimodal deep sensor fusion for object detection, detection of buildings in aerial images, object detection in snow, obstacle detection in thick vegetation using lidar, and evaluation of mixed real and synthetic datasets. We will discuss MAVS simulation of environmental effects such as rain, snow, and fog, and how they contribute to generation of synthetic camera and lidar data.

Links and sources

• Official / arXiv page

Comments