How to work with few features?

Hi,
Working through the reinforcement learning tasks from denny britz (https://github.com/dennybritz/reinforcement-learning) I came across the problem that I had very few features to learn from.

For example in the mountain car example in the openai gym you just get 2 features and have to predict actions.

Anybody experience with these situations? In the solution a RBFSampler from Scikit learn is used and I’ve read something about mapping the features into higher dimensional space (svm sometimes require that)

Are these kinds of preprocessing steps also helpful for neural nets or is it better to just brute force learn with a few dense layers?

Johannes