Aiming to address the challenges of limited operating space, large variations in fruit-cluster height, and complex obstacle distributions in tomato harvesting under vertical cultivation conditions in protected greenhouses, a tomato harvesting robot that integrated visual perception, decoupled motion planning for the lifting mechanism, and a flexible gripping-cutting end-effector was designed and developed. The system adopted a hybrid configuration consisting of a 7-DOF collaborative manipulator and a rail-mounted lifting mechanism, and employed a dual-layer RGB D perception framework combining long-range and close-range sensing to enable accurate detection of ripe fruit clusters and precise localization of pedicel cutting poses. To improve the feasibility of harvesting trajectories under varying height and posture constraints, a decoupled planning method based on a reachability map for the lifting joint and manipulator was proposed. This method was further combined with improved PB PSO inverse kinematics optimization and attraction-field-guided RRT?path planning to achieve fast and collision-free generation and execution of harvesting trajectories. Field experiments in greenhouse environments demonstrated that the proposed method improved the accuracy of cutting-pose estimation and the success rate of grasping-pose planning by 5.9 percentage points and 7.5 percentage points, respectively, compared with baseline methods, achieving an overall harvesting success rate of 65.8%. These results validated the feasibility of the system in complex protected-greenhouse environments and underscored its potential to advance efficient and intelligent protected agriculture.