ADAS Scenario-Based Testing: Transfer Learning from Virtual to Physical Scenarios

The increasing complexity of autonomous driving from basic cruise control to full autonomous driving calls for an efficient testing framework. It is challenging to validate a complex ADAS system considering the diversity of traffic conditions such as road types, complex traffic, weather conditions. Recently, the ADAS industry has relied on scenario-based approach with a V&V-cycle for testing. This process however poses several challenges: 1: simulate realistic scenarios virtually that are beneficial for physical testing; 2: when transferring from virtual to physical scenario-based testing we still need do a great number of real tests. 

Today, The Simcenter Autonomy team presents an efficient ADAS scenario-based testing process, with focus on how to reconstruct and adapt virtual test cases via exploiting real logged vehicle sensors data (camera, lidar, GPS) over a sensor fused perception layer that helps to make the scenario more realistic and close the gap between virtual and real test cases.