Robotic manipulators are a promising alternative to manual pruning in orchard management. Rather than developing crop-specific robotic systems, recent advances in commercial manipulators enable a software-driven approach. However, transferring robotic pruning from laboratory settings to real orchards remains challenging due to the environments with cluttered obstacles, uncertainties, and limited mechanical accuracy. Previous solutions normally assume a robotic system with accurate calibration and no uncertainties. However, building such a system is costly, therefore blocking the lab-to-field transfer. In this work, we present a planning and control system for robotic pruning based on a dual-modal motion framework. The system integrates global motion planning with local reactive control to balance safety and execution accuracy under uncertainty. The key components include: (1) a region-based motion planner that relaxes strict target constraints, (2) a conjunction mechanism that is robust to uncertainty, and (3) a reactive control module based on point cloud alignment. The overall system also incorporates a dual-camera perception setup and a pruning command generator, forming a complete end-to-end workflow. The proposed system is evaluated in three stages: simulation, an indoor laboratory setup, and deployment in a real apple orchard. All experiments are conducted with either real orchard trees or the data collected from a real orchard. For the field experiments, the system is built from accessible components to enable rapid prototyping, including a commercially available robotic manipulator and a standard agricultural tractor. The results demonstrate that the system can be effectively applied in real orchard environments, providing improved robustness and performance. Finally, we release an open-source dataset containing point cloud data and cutting commands collected from the orchard experiments.