| Spiking Neural Networks are receiving a lot of attention as a powerful, computationally efficient, and hence energy efficient alternative to traditional artificial neural networks. In this paper, we present a new method to solve the lunar lander problem using a deep Q-learning approach with an SNN. Our contribution is a new spike encoding algorithm to encode the state vector. This makes it possible to achieve a good result with the SNN in only one simulation step. In addition, we have tested several input layer sizes and network architectures. Our results are evaluated in terms of performance and runtime on a GPU. Our SNN with one-hot encoding outperforms the state-of-the-art ANN with respect to its hardware requirements while still reaching a comparable accuracy. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.