Exploiting plant dynamics in robotic fruit localization

The task of harvesting a tomato truss requires the truss to be separated from the stem by severing the peduncle. Challenges to perform this task with an autonomous robotic system include ripeness detection of the truss and localizing the peduncle, which is often done using only a vision system. Occlusion by leaves, stems and other trusses generally hamper vision-based detection. This study proposes a supplementary method to reach the peduncles, based on vibration feedback. This is done by mechanically exciting the tree at the stem and measuring the frequency response. At a certain frequency, the truss will resonate to the excitation. When traveling along the stem, while exciting it, the shift of this resonance frequency indicates the progress towards the peduncle. It is demonstrated that the resonance frequency increases when the distance to the peduncle decreases, which is attributed to a change of the mechanical stiffness of the stem part between the location of excitation and the location of the peduncle. The method is first tested by exciting a mock-up tomato plant with a 1 degree-of-freedom (DOF) actuator and measuring the response. This mock-up behaves like a fourth order mass spring damper system, where the stiffness of one of the modeled springs increases when the distance to the peduncle decreases. Next, the method is tested on a real tomato plant with multiple trusses. It is shown how these trusses contribute to the multiple distinguishable resonance frequencies. The individual contribution of the trusses to the response can be identified due to the superposition principle for linear dynamics. The closest truss has the highest resonance frequency of 3-4Hz. Similar to the mock-up plant, it is shown that these frequencies increase when approaching the peduncle, with approximately 0.7Hz/m. The change of the frequencies should be used to track the progress towards a truss, rather than predicting an exact location. This approach is independent of the exact parameters of the plant, resulting in a promising proof-of-concept as extra sensor modality.