Autonomous dump trucks in mining sites face significant challenges in path planning due to complex terrain characteristics, variable operational conditions, and stringent energy efficiency requirements. This paper proposes a novel hybrid algorithm integrating Adaptive Hybrid A* with Risk-Aware Reinforcement Learning (RA-RL) to address these challenges. The method leverages terrain-aware cost maps, comprehensive vehicle dynamics modeling incorporating rolling resistance and acceleration factors, and systematic reinforcement learning with proper action space discretization to optimize energy consumption, safety, and efficiency. A rigorous mathematical model incorporates realistic energy cost functions and bicycle model kinematic constraints. Experimental validation using diverse terrain scenarios and static obstacle configurations demonstrates superior performance compared to state-of-the-art methods, achieving 12% energy reduction and 18% computational efficiency improvement with systematic component validation and statistical significance confirmation.