The increasing trend of deploying advanced artificial intelligence systems in real-world applications has raised the need to ensure the robust alignment of artificial intelligence with human values and safety. However, the traditional reinforcement learning mechanism relies on the concept of predefined reward functions, which may not accurately capture the complex human intention. This may lead to the emergence of reward mis-specification and unsafe decision-making. To overcome these challenges, a unified framework is proposed in this paper for the robust superintelligence alignment of artificial intelligence. The proposed unified framework is based on the integration of human value learning, constrained reinforcement learning, and explainable artificial intelligence. The proposed mechanism is based on the concept of reinforcement learning from human feedback, in which the human preferences are learned. In addition, the proposed mechanism is based on the concept of optimization constraints, in which the safety of the decision-making process is ensured. Furthermore, the proposed mechanism is based on the concept of explainable artificial intelligence, in which the decision-making process is explained. The proposed mechanism is evaluated on a human preference-based dataset. The results of the proposed mechanism show that the proposed unified framework is effective in aligning artificial intelligence with human values. The proposed mechanism is applicable in real-world applications. The results of the proposed mechanism show the significance of integrating human value learning, constrained reinforcement learning, and explainable artificial intelligence in the decision-making process.