A Survey of Environmental Perception for Unmanned Ground Agricultural Machinery in Field Environments

Unmanned ground agricultural machinery is required to operate efficiently in complex and dynamic field environments, which presupposes accurate and reliable environmental perception capabilities. This requires the machinery to perceive and respond to various typical elements in both driving and operational environments, such as obstacles, crop rows, and field boundaries. This paper focuses on typical environmental elements and analyzes the environmental perception technologies used in unmanned ground agricultural machinery during field navigation and operation. First, the working principles, advantages, limitations, and application scenarios of commonly used sensors, including vision and radar sensors, are comprehensively reviewed. In addition, the critical role of multi-sensor fusion in enhancing perception robustness and adaptability is highlighted. Subsequently, this paper centers on the specific environmental elements encountered by unmanned ground agricultural machinery. From this perspective, existing perception methods are systematically categorized and reviewed across three domains: image data, point cloud data, and multimodal data fusion. The performance differences and applicable scenarios of these methods in practical applications are also analyzed. Finally, the current challenges facing environmental perception technologies for unmanned agricultural machinery are analyzed, including multi-sensor fusion complexity, the computational-real-time trade-off, and the scarcity of specialized datasets. Future development trends and potential research directions are also discussed. This review aims to provide a reference and foundation for advancing environmental perception technologies in unmanned ground agricultural machinery.