The application of greenhouse gases quotas has led the automotive manufacturers to increase the electrification level of their vehicles. In parallel with Battery Electric Vehicles (BEV), hybridization solutions have been developed. Among them, mild-hybrid technology allows the connection of an electric powertrain with an Internal Combustion Engine (ICE) with the aim of absorbing peaks of fuel consumption. In order to remain competitive, the manufacturing costs of a vehicle need to be optimized. In that regard, removing the stator currents sensors allows avoiding their inherent costs. However, within the vector control framework, a feedback on these currents is required to optimize their value for a given torque. For this reason, it has been decided to use state observers to estimate the missing currents. Different state observer solutions have thus been developed: the Extended Kalman Filter (EKF) and a state observer with two extensions whose design is based on a convergence analysis using Lyapunov functions. With the aim of improving the precision of the stator currents estimation, an in-depth study of the machine's electrical model was carried out. It allows minimizing errors due to parametric variations, related in particular to the magnetic saturation of the machine and uncertainties due to unmodeled phenomena in the whole drive. A method for mapping the machine was proposed using a parametric estimator. The experimental results, obtained on a test bench built in the laboratory, are conclusive in steady-state: the real currents are estimated with a satisfying precision for an automotive application and allow performing a current sensorless control of the machine.