Real2Sim Closed-loop Framework in Agriculture Robotics: A Review

Robotics has profoundly advanced off-road automation, yet developing robust field systems remains challenging due to environmental variability, limited data, and extensive real-world testing. These challenges are especially critical in agriculture, where seasonal constraints and the risk of crop damage limit large-scale experimentation. Simulators and Sim2Real techniques have been introduced to mitigate these difficulties, but they struggle to close the gap between simulated and physical domains because of insufficient realism. Real2Sim closed-loop framework emerges as a promising paradigm to facilitate bridging the gap, which extends simulation-based approaches by constructing high-fidelity simulation from real-world data and establishes a bidirectional data flow between two worlds. This survey provides a comprehensive review of Real2Sim closedloop framework in agricultural robotics. We first establish the three core components of the framework: Real2Sim, Sim2Real, and feedback mechanisms. We then thoroughly examine agricultural robotics literature with particular emphasis on Real2Sim methods, categorizing approaches into procedural, physics-based, and data-driven strategies. We further analyze how Sim2Real techniques address distinct requirements for perception and control applications, and discuss simulation fidelity evaluation approaches in agricultural contexts. Finally, we identify persistent challenges and outline promising directions for advancing Real2Sim closed-loop framework and its applications in agricultural robotics.