Trajectory Optimization for Falsification: A Case Study of Vehicle Rollover Test Generation Based on Black-box Models

In this paper, we consider optimization of trajectories for automotive vehicle rollover testing. In particular, worst-case trajectories that are most likely to cause rollover accidents are determined through trajectory optimization. Our approach combines online local-model identification and gradient-based input update, and can be applied to black-box type models, e.g., a high-fidelity vehicle dynamics model given as a simulation code and not as an explicit set of equations. With our approach, a library of worst-case trajectories corresponding to different operating conditions (e.g., vehicle mass, road surface conditions, etc.) can be constructed and subsequently used in hardware tests. Copyright (C) 2020 The Authors.