Abstract: With the continuous advancement of agricultural modernization and intelligent farming, agricultural robots have increasingly been recognized as a key technological approach for improving agricultural productivity, alleviating rural labor shortages, and supporting the development of smart agriculture. Compared with structured industrial environments, agricultural working scenarios are characterized by strong openness, high dynamics, and pronounced unstructured properties. Large variations in crop morphology and growth stages, complex and uneven terrain conditions, as well as multiple sources of environmental uncertainty such as illumination changes and weather fluctuations, significantly increase the difficulty of robotic perception, decision-making, and control. As a result, the research, testing, and deployment of agricultural robots often suffer from long experimental cycles, high costs associated with field trials, limited availability of large-scale real-world data, and insufficient repeatability of algorithm validation. These challenges severely restrict the efficiency and scalability of agricultural robot development. Against this background, virtual simulation technologies have gradually become an important enabling tool for agricultural robotics research and development due to their advantages in safety, cost controllability, and experimental repeatability. Virtual simulation provides a controllable and reproducible environment in which robotic systems can be designed, tested, and evaluated under diverse operating conditions without the risks and expenses associated with real-world agricultural experiments. In recent years, advances in computing power, physical modeling, and graphics rendering have further promoted the adoption of high-fidelity simulation platforms in robotics research. This paper presents a systematic review of the development of virtual simulation technologies, tracing their evolution from early applications in military training simulators to modern high-fidelity and multi-modal simulation systems that integrate computer graphics, physical modeling, artificial intelligence, and interactive technologies. Furthermore, the application trajectory of virtual simulation in the robotics field is reviewed, with particular emphasis on its expanding role in agricultural robotics research. On this basis, the major application directions of virtual simulation in agricultural robot development are summarized, including modeling of unstructured agricultural environments, kinematic and dynamic analysis of robotic systems, verification of operational workflows and task strategies, safety evaluation of human-robot collaboration, and operator skill training. Focusing on representative agricultural operation scenarios, this paper reviews recent studies conducted using mainstream simulation platforms such as Gazebo, Unity3D, and NVIDIA Isaac Sim. The applicability, advantages, and limitations of these platforms at different stages of agricultural robot development are analyzed, particularly in terms of physical simulation accuracy, visual realism, interaction capability, and integration with robotic middleware systems. By examining typical agricultural tasks such as orchard harvesting, crop protection, pruning, and seeding, this review summarizes application examples of domain randomization, computer vision algorithms, autonomous navigation methods, and procedural content generation within virtual simulation environments. These examples demonstrate the supporting role of simulation in reducing real-world experimental risks, accelerating algorithm iteration, and improving development efficiency. In addition to its advantages, the application of virtual simulation in agricultural robotics still faces several key challenges. These challenges include difficulties in environment modeling and task adaptability due to the complexity and variability of agricultural scenes, intrinsic technical limitations and application bottlenecks of simulation systems, and insufficient accuracy in modeling physical interactions and contact dynamics between robots and agricultural objects, such as crops and soil. Moreover, discrepancies between simulated and real-world sensor models and environmental conditions, further exacerbate the gap between simulation results and actual robotic performance. Finally, this paper discusses future development trends of virtual simulation technologies in agricultural robotics. With continuous progress in high-fidelity, multi-scale, and multi-modal simulation, as well as deeper integration with artificial intelligence, digital twin technologies, and data-driven modeling methods, virtual simulation is expected to play an increasingly important role in research tasks of agricultural robots, such as algorithm validation, system evaluation, and methodological comparison. Overall, this review aims to provide a comprehensive reference for researchers seeking to apply and further develop virtual simulation technologies in agricultural robotics and smart agriculture.