Psychological Restrain for Fault Detection in Reinforcement Learning

Reinforcement learning is a sub part of machine learning that aims for adapting its agent or agents to the unfeasible environment with a solo purpose of achieving the maximum rewards. In common RL agents are trained to find the optimum policy for the varied actions and to achieve the maximum rewards possible. Markov Decision processes that call a rule of penalty if and only if it got the penalty or transit to achieve the penalty. But however, getting all the data set values that shows all those transit states or chances where it couldn't be able to get the rewards. In this paper, we put forward the Reinforcement Learning psychological restrain for fault detection that will visualize the restrains that block the agent to optimize the rewards. These bugs can be in the logical part of the code or there might be a misplaced value of rewards or penalty that got missed by the programmer. Its effectiveness can be shown via snake game.