The aerodynamic drag of cars and trucks plays an important role for energy efficiency affecting travel range and operating costs. This drag can be significantly reduced by actuators ranging from passive to closed-loop active devices. A key feature, opportunity and technical challenge is the inherent nonlinearity of the actuation response [1]. For instance, excitation at a given frequency will affect also other frequencies. This frequency cross-talk is hardly accessible in any linear control framework. The challenge is amplified when employing multiple actuators and sensors as well as multiple operating conditions. Recently, Artificial Intelligence (AI) / Machine Learning (ML) has a opened game-changing new avenue [2]: the automated model-free discovery and exploitation of unknown nonlinear actuation mechanisms directly in the plant. In this talk, we present recent advances of machine learning control [3, 5] for car and truck models with multiple actuators and sensors in experiment and in simulations. Examples include 22% drag reduction of a square-back Ahmed body with feedback control, 17% drag reduction of slanted Ahmed body with 10 actuation parameters [4], and the learning of drag / side force control of a truck model during transients. We show that even complex control laws can be optimized in surprisingly short learning times.