Abstract: Orchard inspection robot has widely been one of the most important equipment in modern orchard management using information technology. The robot with walking chassis and long arm is more suitable for this work than UAV, in cases where long-term monitoring is required, especially when it must be used in severe weather conditions. But the arm body is prone to shake, when the robot is walking in wind and rain or on uneven ground. If the long arm structure with strong rigidity is used, it will cause the overall bulkiness and the danger of overturning the whole vehicle easily, because of the high center of gravity. Therefore, the low stiffness arm body is beneficial to the safety of vehicle. Nevertheless, the lightweight long flexible arm tends to low-frequency vibration, due mainly to the low stiffness. Such low-frequency vibration can also cause another trouble, that is, high-quality image information cannot be captured during the vibration, and the process of waiting for vibration to subside is too long, which seriously restricts the acquisition efficiency of image information. If the stability control can be realized, this flexible characteristic will be conducive to the overall stability. Alternatively, Finite Element Method (FEM) has widely been one type of reliable calculation, when the analytical dynamic model cannot be obtained, due to the existence of concentrated mass or non-single section in the long flexible arm. FEM modal analysis can be utilized to greatly simplify the control model, where the cantilever part of long flexible arm was equivalent to a two degree of freedom pseudo rigid body. Particularly, this simplification can be carried out under the condition of low-frequency vibration. Therefore, this study aims to effectively suppress the long-flexible arm shaking of orchard inspection robot, further to improve the efficiency of image acquisition. Correspondingly, a vibration suppression control system was also proposed using the synthesis and feedback of elevation angles from three parts of arm. Firstly, a dynamic model of equivalent three-bar two torsion spring was established using the FEM modal analysis on the external extension of arm, where the equivalence of natural frequency and vibration modes depended mainly on the static equilibrium. The readings of three inclination sensors were then synthesized into a system output, according to the Differential Flatness theory. As such, the significant vibration of long flexible arm was rapidly suppressed within 9 s under the control of differential flat output as feedback. Specifically, the amplitude at the end of arm reached 10° decreasing to less than 2° within three control cycles, but the small amplitude and high frequency vibration were difficult to eliminate in the later stage, particularly when the PID controller was used. Fortunately, the profile of system output was relatively soft, and the curved of torque output was saturated twice, significantly less than that of PID, when ADRC controller was adopted. Although large vibration was effectively suppressed after five cycles, it was not easy to occur high-frequency jitter in the later stage. This control system can be expected to serve the long-flexible arm mechanism with impact disturbance or arm body jitter under moving conditions, particularly where the active vibration suppression is needed