AI-based perception and prediction of a critical event as a first step for shadow mode testing of the ACC function
摘要
We motivate the development of a shadow testing mode for the ACC function and realize a recurrent neural network that recognizes certain standard scenarios and predicts a critical event in each of these scenarios. This is a first step in setting up the entire AI-based shadow system which, in addition to the perception and prediction stage for edgy scenarios, comprises an automated comparison between simulated predictions of the actuators for a small temporal neighborhood of the critical event in actual driving and in shadow mode.