Reinforcement Learning (RL) has emerged as a powerful paradigm for addressing challenging decision-making and control tasks. By leveraging the principles of trial-and-error learning, RL algorithms enable agents to learn optimal strategies through interactions with an environment. Over the years, RL has achieved remarkable successes in various domains, ranging from game playing to robotics and beyond. However, despite the success of these RL algorithms, their practical application in the real world still faces several challenges such as sample inefficiency and the lack of interpretability arising from the reactive nature of RL policies. RL algorithms require a substantial number of interactions with the environment to learn effective policies. This limitation hinders the applicability of RL in scenarios where data collection is expensive or time-consuming. In some environments, this data collection can also be potentially unsafe. Interpretability is crucial for understanding and trusting the decisions made by RL agents. RL algorithms are regarded as black box reactive policies, making it challenging to interpret the decision-making process, understand the factors influencing agent behavior, or even carry out safety checks before executing any commands. The impact of these challenges are aggravated in safety-critical environments, which includes many application areas within robotics. This work tries to address these challenges in the practical application of RL, particularly in safety-critical environments involving robotics applications. Overcoming these challenges will facilitate the adoption of RL in real-world settings, enabling intelligent decision-making and control in safety-critical domains.