Abstract
The intensifying global demand for sustainable agriculture has necessitated innovation in weed management, particularly through intelligent, non-chemical alternatives. Among these, smart mechanical weeding systems integrating artificial intelligence (AI), machine vision, and robotics are emerging as transformative tools for precise and eco-friendly weed control. While several recent reviews have examined intelligent weeding or machine vision-based weed management more broadly, a comprehensive and systematically structured synthesis focusing specifically on AI-driven mechanical weeding systems that integrate both vision and robotic actuation remains limited. This study presents a systematic review of 176 technical papers published between 2000 and 2024, with in-depth analysis of 33 key works, aiming to explore the design and performance of intelligent mechanical weed control systems in precision agriculture. The review investigates foundational mechanical weeding methods, recent advances in sensor integration and weed detection algorithms, and the use of robotic platforms for intra- and inter-row weeding. It highlights the critical role of RGB, LiDAR, hyperspectral sensors, and deep learning models in enabling real-time, selective weed removal. Comparative case studies showcase end effectors, control architecture, sensors, and techniques involved across diverse platforms. While significant progress has been made, challenges persist in weed-crop differentiation, model generalization, real-time actuation, and economic feasibility. The review proposes a set of design and operational guidelines addressing sensor fusion, adaptive tooling, platform modularity, and user-centric interfaces. This work provides a targeted, system-level roadmap for researchers, developers, and stakeholders in agricultural robotics, offering insights into current capabilities, gaps, and future directions to advance intelligent mechanical weeding for scalable and sustainable food production.