This article evaluates scattering center-based simulation techniques and raytracing-based radar simulation techniques to train a neural network to recognize human gestures and certain activities. Simulation techniques may be more effective to train neural networks and achieve deep learning than observed datasets because simulation data sets may be far larger than observed datasets. A key challenge is to assure that the simulation datasets are realistic and high quality, and several approaches to achieve high quality datasets are addressed. Domain gaps remain for future research.