Research Review on Flexible Manipulator Picking Based on Machine Vision

Abstract

With the continuous advancement of agricultural intelligence, greenhouse agriculture has become a core area of development, and automated harvesting is a key requirement for improving its production efficiency. However, the complexity and variability of the greenhouse environment and the insufficient adaptability of traditional rigid robotic arms have led to increasing attention to research on harvesting using flexible robotic arms based on machine vision. This paper reviews the key technologies and research progress of machine vision and flexible robotic arms. In the machine vision section, most current work focuses on fruit target detection and localization, improving image algorithms or innovating feature extraction methods, and even introducing deep learning methods to improve the target detection rate and robustness of fruits under different lighting conditions and occlusion conditions. For the flexible robotic arm itself, researchers. Improvements are mainly made from two perspectives: motion planning and control methods, to better cope with the complex working environment and constraints inside greenhouses and minimize damage to plants and fruits during operation. Although progress has been made, many problems in this regard still need to be solved. For example, it is necessary to make the vision system more robust, and it is difficult to dynamically model the flexible robot arm, so the control accuracy is not good enough. In the future, technologies such as multi-modal large-scale models and digital twins can be used to mix information from multiple sensors. This can improve the adaptability of the sensing system and the intelligent control level of the robot arm. If we summarize and summarize what we have found, it can provide insights and guide future research.