Wall building is labor-intensive and time-consuming as a typical construction task. It is expected to perform this heavy task using robots. However, laborers are still largely employed, mainly because high continuous stacking precision for mobile robots has always been a bottleneck problem. This paper examines the characteristics of continuous robotic stacking in wall-building tasks and finds the key challenge lies in precisely manipulating the robot arm terminal to the target pose and guaranteeing the wall’s verticality. We propose an object-based terminal positioning solution for this issue. The core of this solution is establishing a reference coordinate system as a measurement reference and designing a multi-sensor measurement system to measure the relative pose between the robot and the object. The solution allows the robot to reposition multiple times and iteratively correct the terminal’s pose, which greatly reduces the impact of positioning errors and loaded deflections of mobile systems on terminal positioning accuracy. Furthermore, we propose for the first time the construction of an active reference feature using a planar laser as a global constraint to provide a consistent reference for establishing reference coordinate systems. Global constraint eliminates the cumulative stacking errors due to relying only on the previously stacked object’s reference features for terminal positioning in continuous stacking tasks and guarantees the wall’s verticality. Finally, an autonomous mobile stacking robot is developed, and continuous block stacking experiments are conducted with this robot in a laboratory environment simulating real application sites. Experimental results demonstrate that the developed robotic system can achieve millimeter-level stacking accuracy, where a single vertical stack error of 3 mm meets the high requirements of the construction industry. Thus, it is reasonable to believe the proposed solution is extremely competitive compared to existing studies and provides valuable insights for improving the continuous stacking accuracy of mobile robots in large-scale unstructured environments.