The project contributes to efforts to increase the range of electric vehicles by developing an energy-efficient torque vectoring control system for vehicles with direct multi-motor all-wheel drive with disconnect clutches. Two complementary control strategies are proposed and comparatively evaluated: the first one based on Model Predictive Control (MPC), and the second one exploiting Machine Learning/Artificial Intelligence (ML/AI). A high-fidelity forward-looking model of vehicle dynamics and powertrain is first established within a high-fidelity simulation environment, followed by the development of a corresponding computationally efficient backward-looking model, including a set of naturalistic driving cycles for planar vehicle motion. Extensive optimizations of torque vectoring control variables are be performed by applying a globally optimal dynamic programming algorithm to the backward-looking vehicle model over a wide range of driving cycles in order to establish a benchmark and generate a dataset for ML/AI model development. The emphasis of the MPC strategy development is on formulating and computationally efficiently solving of the problem of simultaneous minimization of the energy consumption and the clutch switching frequency, as well as on developing predictive models of vehicle speed and yaw rate using road path preview information. In the development of the ML/AI strategy, the emphasis is on supervised learning based on the optimization results, with reinforcement learning examined as a potential alternative. The proposed control strategies are implemented and tested on a processor- and driver-in-the-loop experimental setup.