This thesis presents a novel approach to address the challenges encountered by a mobile manipulator engaged in dynamic locomotion within cluttered environments. The proposed technique involves the use of a dual monocular visual-inertial odometry (dual-VIO) strategy, which integrates two independent monocular VIO modules, one at the mobile base and the other at the end effector (EE). These modules are intricately coupled at the low level of the factor graph to provide a robust solution. The approach leverages arm kinematics to treat each monocular VIO as a positional anchor in relation to the other, thereby introducing a soft geometric constraint during VIO pose optimization. This mechanism effectively stabilizes both estimators, mitigating potential instability during highly dynamic locomotions. The performance of the proposed approach has been rigorously evaluated through extensive experimental testing, directly comparing it to the concurrent operation of independent dual Monocular VINS (VINS-Mono). The envisaged impact extends beyond the specific application, as the approach may lay the groundwork for multi-VIO fusion and enhanced system redundancy within the realm of Active-SLAM (A-SLAM).