Formal Analysis of a Neural Network Predictor inShared-Control Autonomous Driving
John Grese, Corina S. Păsăreanu, Erfan Pakdamanian · AIAA Scitech 2021 Forum · 2021
View Video Presentation: https://doi.org/10.2514/6.2021-1580.vid Autonomous driving systems may encounter scenarios where it is necessary to transfer control to the human driver, for instance when encountering unpredictable dangerous road conditions. To be able to do so safely, the autonomous system needs an estimate of how long it will take for the human driver to take control of the vehicle. Deep neural networks can be used for making such predictions, however proving that neural networks meet critical safety requirements presents a challenge. We present a formally verified neural network which predicts "Takeover-time" in a shared-control autonomous driving system. The network is trained on data collected from a (semi-)autonomous driving simulator. We use Marabou (a formal verification tool), to analyze the network’s sensitivity, local robustness, contextual robustness, and to find adversarial inputs which produce unsafe outputs.