Currently simulations of crash tests, as physical crash tests, are based on “lab” environment, prescribed scenarios, and regulations. Load-cases and boundary conditions are carefully and specifically defined to “assure” safety to the occupants and pedestrians up to acceptable limits, and to certify the vehicle for road use. Although these design methods and safety protocols have been constantly improving vehicle safety, no “real-case” crash scenarios are being tested, which could contribute to the improvement of the vehicle’s safety. In the presented work, “real-case” data from an autonomous driving software were used for the crashworthiness simulation, beyond the regulated scenarios, with the employment of Machine Learning. CARLA software is a simulator for autonomous driving that can reconstruct and simulate real-world traffic accident scenarios, with various vehicle types, and provide pre-crash data such as: speed, position, and angle. Such data from reconstructed accident scenarios are used as input in Finite Element Crash analysis to provide results concerning occupant injury. Datasets with various FE models and various crash scenarios measuring occupant safety are created to train Machine Learning models that will be able to predict the occupant injury, having as input the pre-crash data of a simulated accident scenario. These Machine Learning models are used to optimize the control of occupant safety parameters such as, airbag deployment time, and seatbelt trigger, to achieve lower occupant injury criteria. In this study, CARLA provided data of a specific vehicle on a rear crash scenario, one of the most common accidents. The input parameters were speed, velocity, and the relative position of the two vehicles. FE analyses run for several variations of the crash, measuring the occupant injury, and creating an adequate dataset to train a Machine Learning model. The trained Machine Learning model was then used to predict the occupant injury criterion based on various inputs from CARLA, and to optimize the safety systems parameters to achieve safer designs.