High-precision localization is fundamental to autonomous driving. However, the complex underground coal-mine environment severely attenuates global positioning system (GPS) signals, resulting in poor localization accuracy for the underground unmanned transport vehicle. To address this issue, we propose a high-precision localization method for underground unmanned transport vehicles in GPS-denied environments by integrating multiple miniature inertial measurement unit (MIMU) with differential heading and average wheel-speed constraint. By fusing measurements and constraint conditions from multiple low-cost MIMUs, the proposed method effectively enhances vehicle localization accuracy in underground environments. First, a novel distributed multi-MIMU architecture consisting of one body-mounted MIMU and two wheel-mounted MIMUs is proposed, and the system error-state model is formulated based on the simplified PHI-angle error theory and a first-order Gauss-Markov process. Secondly, using measurements from the left and right wheel MIMUs, two measurement error models, namely differential heading and average wheel speed, are formulated. Based on these models and the predicted states, the corresponding observation residuals and observation matrices are further derived. Finally, a centralized error-state Kalman filter framework with the innovation-based adaptive filtering is designed to fuse body and wheel MIMU information, while an abnormal measurement detection mechanism is developed. Together, they realize joint estimation of the vehicle pose and effectively improve localization accuracy. Localization experiments using a four-wheel differential-drive robot in three different scenarios show that, in the GPS-denied underground environment with narrow shafts, the unmanned transport vehicle achieved RMSEs of 0.722 m in position and 0.835° in heading. Compared with using only one MIMU, the positional RMSE improved by 1~2 orders of magnitude, reaching a level comparable to GPS-only positioning. Overall, the proposed method exhibits strong drift suppression and stable localization performance, highlighting the substantial potential of low-cost MIMUs for unmanned transport vehicle localization.