Microscopic traffic models (MTMs) are widely used for assessing the impacts of autonomous and connected autonomous vehicles (AVs/CAVs). These models use car following (CF) and lane changing models to replicate the AV and CAV driving behaviors. Several studies attempt to replicate the accurate configuration of these behaviors (especially CF behavior) with many state-of-the-art modeling methods. However, they need to define certain parameters either based on assumptions or estimation by trajectory data from the limited field experiment of AVs and CAVs, and the impacts prediction accuracy depends on the definition of these parameters. For human-driven vehicles, these parameters mimic human drivers, whereas, for AVs and CAVs, most of these parameters could be controlled by an agent (AV and CAV). Therefore, it is possible to train AVs and CAVs to behave in a way that could potentially enhance their related impacts, e.g., traffic efficiency, emissions, and safety. Thus, this paper proposes an optimization framework that tends to find sets of optimized driving parameters for AVs and CAVs under different varying scenarios to achieve pre-defined policy targets (e.g., reducing travel time, number of conflicts). The proposed framework comprises an optimization module and a simulation environment. The differential evolution (DE) method is used within the optimization module to find the optimal values of the CF parameters. The simulation environment is a SUMO-based platform where several simulations are run under certain scenario conditions. An experimental setup is designed to apply the proposed framework under different scenarios of mixed traffic and demand situations. The findings of this study reveal that safety could be potentially improved by optimized values of CF model parameters. For each policy, where higher weight is allocated to safety, generated optimized parameters significantly improve safety as well as efficiency.