Abstract: The tomato harvesting manipulator is apt to work in a complicated and unstructured production environment. The research on the trajectory tracking control is a key task. The tomato harvesting manipulator studied in the paper is a redundant manipulator with 7-DOF including two prismatic joints and five revolute joints, which is a multivariable nonlinear system. It is difficult to obtain its accurate dynamic model during control due to the external disturbance and the nonlinear friction, etc. The traditional control algorithm based on this model has poor robustness and cannot achieve global asymptotic stability. A fuzzy logic system can approximate a nonlinear function with arbitrary precision, which has more and more application in the manipulator control. To realize trajectory tracking with high accuracy and stability, a control method based on compute torque-fuzzy logic compensation is proposed to control trajectory tracking for a tomato harvesting manipulator. This method compensates the uncertain part of the dynamic model of the tomato harvesting manipulator via an adaptive fuzzy logic system, and parameters of the fuzzy compensator are adaptively adjusted by using a tuning algorithm derived from the Lyapunov stability theory. The dynamic model of the manipulator is set up based on a Lagrange method. In addition, the uncertain part of each joint in the model is approached by a separate function in order to reduce fuzzy rules and improve the real-time control. To realize universal approximation, the joint deviation and deviation rate membership function is defined as a Gauss type function. The trajectory tracking controller is designed to include the compute torque controller and self-adaptation fuzzy compensation controller. At the same time, the virtual prototype of the manipulator is structured via adding constraints and drives based on a modular design method. A co-simulation platform is established by ADAMS and MATLAB, which consists of a control program module, virtual prototype module, trajectory input module, and display module, etc. The trajectory tracking control system is simulated on the tomato harvesting manipulator using the compute torque-fuzzy logic compensation method and the computed torque method respectively. The trajectory tracking error and the force (torque) output of each joint are analyzed. The results show that the trace tracking average error from the 1st joint to the 7th joint by the computed torque method are 2.238×10-3m, 0.0242m, 0.0132 rad, 0.0526rad, 0.113 rad, 0.1075 rad and 0.0388 rad, while they are 6.65×10-4m, 1.278×10-3m, 0.0131rad, 0.0135 rad, 0.0116rad, 0.0146 rad and 0.0127rad by the compute torque-fuzzy logic compensation control. The control accuracy from the 1st joint to the 7th joint are increased by 70.29%, 94.72%, 0.61%, 74.29%, 89.75%, 86.41%, 67.14% respectively. The trace tracking errors in the compute torque-fuzzy logic compensation vary smoothly with a rapid convergence of the position error. The joints can reach the desired trajectory within 2-3s. The fuzzy compensation force (torque) of each joint varies smoothly with no significant change. The starting force (torque) is highest in the prismatic joint 2 and the revolute joint 4 when the initial errors are the largest, which are 453.127N and 98.33N·m..The force (torque) output of the 1st joint to the 4th joint which are close to the foundation of the manipulator is relatively larger than the other three joints. The torque output becomes more and smaller while nearing to the end-effector. The result provides a reference for motor choosing of each joint. In addition, the whole output force (torque) of each joint is stable and regular without severe vibration during the whole control process. The compute torque-fuzzy logic compensation control method can improve control accuracy and has great robustness, which will lay a foundation for further control study of the tomato manipulator.