Statistics for Robust Reinforcement Learning for Linear Temporal Logic Specifications with Finite Trajectory Duration