Reinforcement Learning on Tensorflow without Gym

Question

I am currently trying to create a simple ANN learning environment for reinforcement learning. I already did fitting via neuronal network to substitute a physical model for a