Challenges in Real-World Deployment of Agricultural Robots: A Systematic Review of Perception, Navigation, and Environmental Adaptation

Agricultural robots have shown strong results in laboratories and controlled trials, yet few systems have reached sustained commercial use on farms. This systematic review examines why performance degrades when robots move from controlled settings to real fields, focusing on three coupled capabilities: perception, navigation, and environmental adaptation. Following the PRISMA 2020 guideline, peer-reviewed journal articles, conference papers, and reviews published between 2020 and 2026 were searched in Scopus and IEEE Xplore, and 38 studies were included in a qualitative thematic synthesis. Perception remains limited by occlusion, variable natural illumination, depth-sensor degradation in sunlight, and the scarcity of diverse annotated datasets. Navigation is constrained by the loss of satellite positioning under canopies, irregular crop rows, and uneven terrain, which has motivated vision-based and learning-based guidance. Environmental adaptation is hindered by seasonal changes in crop appearance, weather, and crop and cultivar variability, so systems tuned for one site often fail at another. Field evaluations of harvesting robots confirm that success rates fall sharply in unmodified crops. Beyond technical issues, cost, safety, legislation, and farmer acceptance shape adoption. The review identifies research priorities, including multimodal sensing, robust localization, adaptive learning, standardized field benchmarks, and co-design of crops and robots.