Progress 09/01/24 to 08/31/25
Outputs
Target Audience:Research findings have been disseminated via various presentations given by the team members (e.g., Karydis gave an invited presentation to an agricultural robotics workshop organized as part of the 2025 IEEE ICRA conference)and unofficial discussions with stakeholders. The overall project scope and reach have also been communicated to the communities of interest via the above means. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?[UCR]A total of two PhD students (Dimitris Chatziparaschis andJingzong Zhou) and twopostdoctoral researchers (Zhichao Liu and Caio Mucchiani) have been involved in aspects of this project. The project has provided them with training and professional development opportunities related to: 1) hardware design, rapid prototyping, and iterative fabrication; 2) robot motion planning and visual sensing algorithm develpment, and integration of artificial intelligence; 3) presenting research findings to a wide variety of audience ranging from other researchers to growers; 4) interaction with stakeholders. Onepostdoctoral researcher obtained his first faculty position(Liu). [UF]We have recruited a student, Samuel Aregbesola, who started his Ph.D. program in the Department of Soil, Water, and Ecosystem Sciences at the University of Florida in the Spring semester, 2025. Samuel Aregbesola, originally from Nigeria, recently graduated from the Department of Agricultural Civil Engineering at Kyungpook National University in Daegu, South Korea. Samuel is supervised by Nikolaos Tziolas and will work under Objective 2, leading pilot activities on behalf of UF. He will be responsible for developing new Edge AI techniques for agronomic tasks, combining his expertise in machine learning and agricultural engineering. [AUTH]In terms of management, we have signed contracts with three new members (Dimitrios Goutzikostas, Chemist M.Sc., Vasilis Rousonikolos, Chemist MBA, Paraskevi Chantzi, Civil Engineer Ph.D.) of the research team. These additions will strengthen the team's human resources in terms of knowledge and contribute to field data collection. Their integration into the project team ensures a broader capacity to address the complex challenges of multi-crop data collection. The expansion of the team is expected to accelerate the progress of the research, while providing greater flexibility and resilience in the distribution of tasks among the various project activities. Furthermore, procedures have been initiated to restructure the scientific team. How have the results been disseminated to communities of interest?Research findings have been disseminated via various presentations given by the team members (e.g., Karydis gave an invited presentation to an agricultural robotics workshop organized as part of the 2025 IEEE ICRA conference)and unofficial discussions with stakeholders. The overall project scope and reach have also been communicated to the communities of interest via the above means. What do you plan to do during the next reporting period to accomplish the goals?The next reporting period will include further research to establish language-guided AI-based aerial robot mechanism design and will initiate the development of synthetic agents and digital twins to evaluate the preliminary feasibility of the AI-based designs. We will also continue performing more field data collections and analyses.
Impacts
What was accomplished under these goals?
In this reporting period, we focused on both objectives of the project, with contributions from each site detailed below. [UCR] Wedesigneda bimanual aerial robot that employs visual perception and learning to detect, reach, and harvest avocados autonomously. The dual-arm system comprises a gripper and a fixer arm. In earlier work, we have shown that applying a rotational motion is the most mechanically efficient way to detach the avocado from the peduncle; however, the peduncle may store elastic energy, preventing the avocado from being harvested. The fixer arm aims to stabilize the peduncle, allowing the gripper arm to harvest. The integrated visual perception process enables the detection of avocados and the determination of their pose; the latter is then used to determine target points for a bimanual manipulation planner. Several experiments are conducted in controlled indoor and outdoor settings to assess the efficacy of each component individually. Further, an integrated experiment in outdoor semi-controlled settings is used for thefeasibility assessment of the overall system. Results demonstrate that all different components can work synergistically to enable robotic avocado harvesting in (semi-)controlled settings. Results also highlight limitations of an airborne harvesting solution and reveal tradeoffs to be considered in the selection of a harvesting robot. This work has appeared in the journal Wiley Advanced Robotics Research. We alsobegan the investigation of AI-guided design of dual-arm systems (work in progress), and an augmented-reality-based interface to control aerial robots in the field (work submitted in Computers and Electronics in Agriculture and currently under review). [UF]The UF team is currently integrating the ULTRIS X20 Premium Hyperspectral Camera 400-1000nm (Cubert, Germany) with a robotic platform (Quadruped Robotics' Robot Dog). The next step involves the integration of a NVIDIA Jetson AGX Orin Developer Kit to enable edge computing applications. Additionally, the hyperspectral data have been tested alongside a PSR+ 3500 spectroradiometer (Spectral Evolution, 350-2500 nm, resampled to Cubert 400-1000nm range) across various reference targets to evaluate its spectral accuracy. Our team also met several times with the UF farm crew and relevant faculty members from the Department of Horticultural Sciences to plan the identification of potential sampling locations in citrus groves across Southwest Florida. Citrus grove in the South West Florida Research and Education Center has been prioritized due to the diversity of citrus trees (e.g., varieties, age, etc.) As outlined in the award agreement, we will conduct object detection tasks (e.g., tree counting) in selected groves and explore soil sampling using hyperspectral sensors under field conditions. Also, fields with exposed soils have been selected for initial testing and data collection. [AUTH]We planned and executed the first part of the field campaign at the project's vineyard pilot at Ktima Gerovassiliou (Epanomi, Greece), where extensive data collection was carried out using advanced imaging equipment. Specifically, a PSR+ 3500 field spectroradiometer (350-2500 nm) and a Cubert FireflEYE 185 hyperspectral camera (in the 450-950 nm range) were employed to capture spectral and visual information from a variety of vineyard components. The sampling covered not only the vines themselves but also grapes, leaves, soil, vineyard poles, and internal pathways. To emulate aerial monitoring scenarios, multiple viewing angles of grape clusters were systematically captured to simulate the drone point of view, thereby enhancing the robustness of the dataset for object detection tasks. In total, approximately 70 high-resolution images were acquired under field conditions. The combination of hyperspectral and Cubert imaging provides complementary datasets, enabling the extraction of detailed spectral signatures alongside spatially explicit visual information. In this context, pretrained YOLOv11 models for grape detection have been applied in preliminary inference experiments on legacy datasets from previous Gerovassiliou field campaigns. These exploratory tests established a baseline for model refinement. The newly collected spectra and imagery will provide complementary ground-truth references for the fine-tuning and validation of AI models, with the ultimate objective of enabling reliable, near real-time grape cluster identification and maturity assessment aligned with the project's technical requirements.
Publications
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2025
Citation:
Z Liu, J Zhou, C Mucchiani, K Karydis, "Vision?Assisted Avocado Harvesting with Aerial Bimanual Manipulation." In Advanced Robotics Research, 2025, pp. 202500003.
|