Development of self-learning powertrains is crucial to deal with complexity and diversity of future ultra-clean and efficient vehicles and to minimize development time and costs. Self-learning powertrains automatically adapt control settings online to minimize overall powertrain efficiency using sensor and preview information. This approach not only offers solutions to deal with system complexity of future ultra-clean and efficient vehicles, but also to minimize the calibration effort by on-road optimization and dealing with unforeseen environmental conditions and system uncertainty.