Edge AI for Smart and Precision Agriculture: An Overview of Robotic Vision for Field Perception and Decision Support

Abstract

Climate change, water scarcity, and labor shortages are increasing the need for autonomous, data-driven agricultural systems. Autonomous robotic platforms such as UAVs and UTVs, as well as mobile cameras, can support this transition by combining embedded robotic vision, multimodal cameras, and edge AI to monitor crops, understand field conditions, and guide timely interventions directly in the field. This paper focuses on autonomous robotic vision as a central enabling capability for smart and precision agriculture. It discusses how RGB, multispectral, thermal, depth, and stereo camera streams can be processed locally to support real-time perception, crop and canopy monitoring, stress and disease detection, field mapping, navigation, and decision support. The paper reviews enabling edge vision techniques, including lightweight deep networks, detection and segmentation models, visual attention, region-of-interest processing, model compression, adaptive inference, and multimodal visual fusion. Representative use cases illustrate how robotic vision can support UAV-based vegetation monitoring, crop stress assessment and precision irrigation, autonomous UTV-based in-field inspection, and edge-enabled decision support. The paper concludes by identifying edge AI device constraints and future directions for robust, secure, adaptive, and scalable robotic vision in agricultural environments.