Hybrid and Learning based Simulation Framework for Radar-LiDAR Sensor Modeling in Realistic Driving Environments

Abstract

This doctoral research focuses on high-fidelity raw data simulation for radar and lidar sensors in autonomous driving applications. The project addresses the critical domain gap between real-world and simulated sensor data across varied Operational Design Domains (ODDs), encompassing diverse ground surfaces, weather conditions (e.g., fog, rain), etc. By integrating physical models, statistical methods, and machine learning architectures, the research ensures that synthetic data remains strictly constrained by the laws of physics. Furthermore, it investigates cross-domain sensor data synthesis (e.g., translating vision or lidar data into raw radar signals) to enhance simulation efficiency and dataset versatility.