Abstract: We propose a framework for training neural networks (NNs) to imitate complex controllers (like Model Predictive Control (MPC)) for control requirements expressed using Signal Temporal Logic (STL). Initially, the NN is trained using data from grid based sampling of an expert controller (e.g., Proportional Integral Derivative (PID), MPC). The trained NN is evaluated against the behaviour specified in STL and retrained with data from the expert controller at states where the NN controller fails. This iterative training continues until there are no failures. Moreover, we introduce a method to evaluate the performance of the learned controller via parameterization and parameter estimation of the STL requirements. We demonstrate our approach with a non-linear flying robot system with MPC.