Exploring various machine learning methods re- vealed that the traditional gradient-based learn- ing algorithms such as the ones used in rein- forcement learning might not be as efficient with certain tasks as they might seem. By drawing inspiration from multiple state of the art ex- amples, the idea of this project is comparing non-gradient-based algorithm such as neuroevo- lution with gradient-based reinforcement learn- ing on an irregular task of training a car to self-drive around circuits with varying complex- ity. The raw quantitative data collected dur- ing evaluation show that neuroevolution is capa- ble of producing solutions to this problem with great speed when compared to the reinforcement learning approach. However, when the reinforce- ment learning approach is allowed to train for long enough, it manages to train models that are outperforming even the ones created by neu- roevolution. Further statistical research is re- quired to see whether the differences in the per- formances are significant.