Abstract: To address the positioning inaccuracies of Real-Time Kinematic (RTK) technology caused by GPS signal attenuation, multipath, and communication interruptions in deep open-pit mines for unmanned mining trucks, as well as the errors in integrated inertial navigation system (INS) and odometer positioning resulting from tire slippage and the accumulation of INS pose detection deviations over time, proposes a “RTK+INS+Odometer” fusion positioning method based on Variational Bayesian-optimized Sage-Husa Adaptive Extended Kalman Filter (VB-AEKF). For precise position detection, a fusion positioning scheme combining RTK global positioning with INS+odometer local positioning is introduced, where the VB-AEKF algorithm estimates and optimizes process and measurement noise covariances in real time to mitigate the impact of system model uncertainties and time-varying noise characteristics on positioning. By integrating INS and odometer data to correct RTK instability, the stability and accuracy of position detection are significantly enhanced. To counteract tire slippage on rugged, vibrating terrain and the accumulation of INS attitude angle errors, a method constructing pseudo attitude angle constraints from displacement information is proposed, effectively suppressing attitude angle drift. A multi-sensor fusion positioning platform for unmanned mining trucks was established, and simulated open-pit mine environment experiments demonstrated that when the RTK signal is obstructed, the maximum positioning errors in the East, North, and Up directions are 0.570 7 m, 0.554 3 m, and 0.364 7 m, respectively, representing reductions of 0.222 1 m, 0.146 5 m, and 0.216 2 m compared to standalone RTK, with decrease rates of 28.01%, 20.90%, and 37.22%. During unblocked turning, the maximum positioning errors in the East, North, and Up directions are 0.008 1 m, 0.009 7 m, and 0.018 6 m, reduced by 0.002 1 m, 0.004 8 m, and 0.006 7 m relative to standalone RTK, corresponding to decrease rates of 8.3%, 20.69%, and 19.14%. On rugged turning sections, the maximum attitude angle error of the MEMS-based INS is 0.312 1°, which is reduced by approximately 20% compared to the error of 0.425 3° without constraints, significantly improving the system's positioning accuracy and stability in complex dynamic environments.