As increasing use of neural networks (NN) and deep-learning (DL) can be observed in various fields, such as voice and image recognition, control, translation, data processing, function approximation, etc. These technologies can also be applied in a specific domain of experimental vehicles. This paper presents and summarizes our recent practical experiences gained during the usage of neural network as sensing and perception subsystem in an experimental vehicle. The paper does not targets general NN problems, only focuses on the narrow subdomain. The current paper also reviews and evaluates the available hyper parameter tunings, optimizers and neural network architectures and proposes tested solutions for enhanced performance. All of the used datasets, source codes and additional materials is available online.