Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach | Read Paper on Bytez