Despite the significant progress of reinforcement learning (RL) in solving complex decision-making problems, its limited interpretability continues to pose a challenge, particularly in safety-sensitive and user-interactive applications. In this chapter, we propose a transparent variant of decomposed reward Q-learning (drQ), which restructures scalar rewards into distinct, interpretable components, allowing the agent to learn Q-values for each reward dimension individually. To enhance the interpretability of decision-making, we introduce two complementary explanation strategies: reward difference explanation (RDX), which clarifies the influence of each component on action selection, and minimal sufficient explanation (MSX), which determines the minimal set of reward aspects necessary to justify a given action. The proposed method is validated in two environments grid maze navigation and CliffWorld that involve trade-offs between multiple objectives. Our results demonstrate that the interpretable drQ model matches or exceeds the performance of standard Q-learning and hybrid reward architecture (HRA), while offering clearer insight into its decision process. Additionally, we simulate human-agent dialogues to assess explanation clarity and usability.