The work aims to address the multi-objective optimization problem of three-dimensional truck loading, so as to enhance loading efficiency, reduce costs, and ensure cargo transportation safety. By integrating mechanical analysis of cargo during transportation turns, the safety region for the cargo's center of gravity was determined, and a multi-objective optimization model for truck loading was constructed. A block-merging strategy was proposed to reduce the decision space. A dual-population constraint-based multi-objective optimization framework based on reinforcement learning and Q-Learning algorithm was designed to improve convergence and solution diversity. Validation with public datasets and case studies showed that, under constraints including the turning center of gravity, the average space utilization rate achieved 92.07%, significantly surpassing other algorithms. In conclusion, the proposed multi-objective optimization algorithm effectively improves the space and load utilization rates for three-dimensional loading problems, offering effective solutions and references for practical applications. Furthermore, the algorithm significantly enhances loading planning efficiency while ensuring the safety of cargo transportation.