Abstract: Robotic manipulators are extensively used in agricultural harvesting operations and represent a pivotal advancement in agricultural automation systems. Harvesting robotic manipulators operate in harsh, unstructured environments such as orchards and greenhouses, where external disturbances and unknown dynamics of the manipulator must be addressed. To enhance the trajectory-tracking precision and disturbance-rejection capabilities, continuous control strategies—particularly adaptive neural network-based approaches—have been widely implemented in harvesting robotic manipulators. However, these conventional continuous control paradigms impose stringent real-time computational requirements; consequently, persistent high-frequency monitoring of system states and manipulator dynamics is necessitated for real-time control computation and transmission, thereby incurring prohibitively high control costs. Hence, this study proposes event-triggered intermittent trajectory-tracking control for harvesting robotic manipulators to balance between control performance and control costs. First, a dual-joint harvesting robotic manipulator dynamic model is rigorously developed based on Lagrangian mechanics principles. This model comprehensively accounts for the inertial properties, Coriolis effects, gravitational loading, and disturbance inputs. To address the challenges inherent in harsh operational environments, a radial basis function neural-network architecture is implemented to estimate the unknown nonlinear dynamics and external disturbances of the system. This compensation mechanism significantly enhances the overall system stability and operational robustness under practical field conditions. Subsequently, a continuous trajectory-tracking adaptive controller is systematically designed using the backstepping control methodology. The second-order nonlinear dynamic structure of harvesting robotic manipulators inherently circumvents the mathematical complexity associated with virtual control derivation, thus rendering backstepping particularly suitable for this application. By applying the supremum-infimum technique, the total operational period is partitioned into actively controlled intervals and dormant rest phases to establish a precisely quantified relationship between control and rest intervals. The implemented strategy operates under dual operational modes: during control intervals, the controller executes event-triggered actions, whereas all control computations and transmissions are suspended during the rest intervals. This control strategy effectively reduces the computational burden and communication bandwidth requirements while significantly extending operational endurance, thereby reducing the overall control cost. Stability analysis employing the Lyapunov theory shows that under the proposed control strategy, both trajectory-tracking and neural-network estimation errors converge to a bounded region, thus enabling the desired tracking objectives to be achieved within the specified error margins. A comprehensive theoretical examination further establishes the uniform boundedness of all closed-loop signals, thus definitively confirming the absence of Zeno behavior (i.e., infinite triggering events within finite time). This critical property is governed by a strictly positive minimum inter-event time threshold, which ensures that no infinite triggering occurs during any finite operational period. Finally, the feasibility and effectiveness of the proposed control strategy are validated through periodic and aperiodic intermittent operation modes. Under an identical total control input, the event-triggered intermittent control strategy with 10% rest intervals achieved a lower controller update frequency by 69.31% compared with continuous control. Meanwhile, compared with the conventional PID control, it improved the overall RMSE for Joints 1 and 2 by 22.18% and 33.63%, respectively. These results collectively verify the strategy’s exceptional ability to achieve an optimal balance among tracking accuracy, computational cost, and communication cost in harvesting robotic manipulators.