This paper presents a novel model-based method for estimating the attitude of underground articulated vehicles (UAV). We selected the Load–Haul–Dump (LHD) vehicle as our application object, as it is a typical UAV. First, we established the involved models of the LHD vehicle, including a kinematic model, the linear and angular constraints of a center articulation model, and a dynamic four degrees-of-freedom (DOF) yaw model. Second, we designed a Kalman filter (KF) to integrate the kinematic and constraint models with the data from an inertial measurement unit (IMU), overcoming gyroscope drift and disturbances in external acceleration. In addition, we designed another KF to estimate the yaw based on the dynamic yaw model. The accuracy of the estimations was further enhanced by data fusion. Then, the proposed method was validated by a simulation and a field test under different dynamic conditions. The errors in the estimation of roll, pitch, and yaw were 3.8%, 2.4%, and 4.2%, respectively, in the field test. The estimated longitudinal acceleration was used to obtain the velocity of the LHD vehicle; the error was found to be 1.2%. A comparison of these results to those of other methods showed that the proposed method has high precision. The proposed model-based method will greatly benefit the location, navigation, and control of UAVs without any artificial infrastructure in a global positioning system (GPS)-free environment.