Navigating Edge Cases with Physics-Based Sensors Simulation for Autonomous Vehicles
Training & evaluating perception systems requires a large variety of accurate & annotated data to ensure safety of Autonomous Vehicle in their Operational Design Domain. Acquiring enough real-life data for sensor perception edge cases or dangerous driving situations is challenging. Synthetic data can be used to fill this gap. But, to obtain trustable data, a physics-based simulation approach should be used for multispectral light and electromagnetic wave propagation in 3D synthetic scene as well as the modelling of the sensor. We will highlight on camera edge cases that are only relevant when being simulated with this approach.