Hoang Minh Pham

Modern viticulture requires precise, consistent pruning to maintain both yield quality and plant health. However, manual pruning is labour-intensive, subjective, and difficult to scale. Automation poses challenges due to the complex shapes of grapevines, dense environments, and limited access for robotic manipulators. Therefore, there is a need for an intelligent robotic system that can efficiently reach and cut pruning targets while ensuring safety and adaptability in cluttered vine structures.

This research introduces a dual-arm robotic pruning system that utilises RGB-D perception and model-based motion planning to enhance manipulability and reachability in challenging vineyard environments. The approach involves several key components: (1) designing and automating reachability tests for multiple robot configurations; (2) integrating these tests with motion planners to ensure complete accessibility; (3) computing optimal pruning poses based on metrics of reachability and manipulability; (4) optimizing pruning sequences to reduce execution time; and (5) developing collision-aware control for maneuvering the end-effectors, guided by depth sensing through an eye-on-hand approach.

This research presents a comprehensive framework for autonomous multi-arm pruning by addressing each stage of the pruning workflow, from reachability analysis to motion execution. The system enhances task efficiency through automated accessibility testing and sequence optimisation, allowing robots to plan and execute pruning tasks with minimal human intervention. It improves manipulation precision and adaptability by selecting pruning poses that offer optimal manipulability and utilising depth-informed feedback control. Additionally, the integration of whole-body collision avoidance ensures safe operation in dense vine environments. Together, these advancements yield higher pruning accuracy, shorter cycle times, and greater overall system autonomy, paving the way for scalable robotic solutions in complex agricultural settings.

Current vineyard pruning solutions primarily rely on manual labour or semi-automated systems that use vision but are limited in adaptability and often employ single-arm kinematics. Many existing robotic manipulators also lack coordinated reachability analysis, collision-aware control, and multi-arm task optimisation. This research goes beyond these methods by integrating reachability-driven planning with sensor-based adaptive control, thereby significantly enhancing autonomy and robustness in dynamic, unstructured agricultural environments.