Robotics and AI-Assisted Disease Phenotyping in Bt Cotton Fields: Transforming Modern Plant Pathology

Abstract

The large-scale use of Bt cotton has greatly alleviated pressure from lepidopteran pests, but this development has inadvertently altered the pathosystem, elevating secondary or opportunistic pathogens as a significant source of losses. Traditional visual estimation techniques for disease assessment reliant on subjective scoring metrics remain costly and inefficient amid large-scale Bt cotton monocultures. In this review, we summarize recent innovations in autonomous ground and aerial robotics coupled with deep learning as a paradigm shift for disease phenotyping in Bt cotton. We discuss the latest advances in robotic technologies; artificial intelligence, including convolutional neural networks and vision transformers; and explainable AI tools for detecting sub-millimeter-level and multispectral plant diseases. In particular, we highlight challenges like disease mimicry in Bt plants and emerging pathosystems, such as Fusarium wilt and Alternaria leaf spots. We also propose a detect-analyze-respond robotic framework that integrates robotics, AI, and molecular techniques for next-generation plant pathology. Finally, we address key bottlenecks, including hardware robustness, generalizability of machine learning models across Bt varieties, and ethical concerns.