Trustworthy agricultural autonomy integrates robot learning safe control and human robot interaction

The future of modern farming is intelligent, autonomous and data-driven farming operations. This will help to address the increasing labor shortage, resource constraints and climate variability. But for deployment in unstructured agricultural settings, high-performance automation is needed, as well as trustworthy systems that can make safe decisions, have the ability to learn adaptively and collaborate effectively with human operators. In this review, we discuss recent progress in trustworthy agricultural autonomy from the perspective of three interrelated pillars: robot learning and perception, safe control and resilient planning, and human-robot interaction. This review surveys state-of-the-art perception technologies such as computer vision, LiDAR, sensor fusion and deep reinforcement learning and safety-oriented approaches including model predictive control, formal verification, failure management and resilient navigation. It also addresses human-robot collaborative systems, explainable AI, ergonomics safety and sustainability-aware robotic design and discusses ethical, regulatory and socio-economic issues related to autonomous farming. Commercial applications of autonomous weeding, robotic harvesting and broad-acre field operations are now in existence, providing real-world demonstrations of the use of these technologies, and their potential to increase productivity, improve resource efficiency and enhance environmental sustainability. Finally, the review wraps up by underlining some future research priorities like universal safety certification, edge computing, interoperability, swarm robotics, and digital twin integration. Collectively, the reviewed evidence identifies the technical, human, and governance requirements that must be addressed to support safer, more resilient, transparent, and sustainable deployment of autonomous agricultural systems. Graphical Abstract