Abstract: System, methods, and other embodiments described herein relate to estimating reachable states for a vehicle using a neural model. In one embodiment, a method includes generating an unverified command by a first controller and a safety command by a second controller for a projected path by an automated driving system (ADS) associated with a vehicle. The method also includes estimating reachable states by a neural network (NN) model that uses the unverified command and an initial condition for the vehicle, the reachable states representing positions and orientations for the vehicle within a bounded time and a state space associated with the initial condition and the projected path. The method also includes, upon determining that the reachable states satisfy collision criteria, executing by the vehicle a maneuver with the safety command for the projected path and generating subsequent commands using vehicle data from the first controller and the second controller.

Abstract: Embodiments of model-based system design with model verification are disclosed. An embodiment includes receiving a model for a system and at least one specification for the system. In some embodiments, the system determines at least one of a minimum expected robustness value and a maximum expected robustness value for a region of a search space of the model with respect to the at least one specification. The model may be modified based on the determined minimum or maximum expected robust ness value.