In order to design a highly efficient apple harvesting robot, an apple harvesting robot was proposed based on gaselectric hybrid that can work around the clock. The design and evaluation of an apple harvesting robot in the laboratory was introduced. The freedom manipulator with five degrees adopted both electric and pneumatic hybrid power as the power source. The electric motor ensured that the arm can accurately position the apple, and the pneumatic motor ensured the endeffector can quickly and flexibly grasp the apple. The apple recognition algorithm combined traditional machine learning methods with the recent popular deep neural network methods to optimize the detection speed while improving the robustness, and used the builtin lighting system to achieve longterm work. The control system included an industrial computer, an AC servo drive, and a pneumatic actuator that controlled the robot arm and the end effector to access and pick the apple. The system carried out visual and picking experiments in a laboratory environment. The average time for visual positioning of apples was 44ms per image. The harvesting system reached an accuracy of 81.25% with average picking time of 7.81s per fruit.