Robotic crop manipulation remains a critical bottleneck due to the complexity of interacting with unstructured biological structures. Because no two crops are exactly alike, reproducible testing is particularly important. While simulation offers a pathway to overcome seasonality and damage risks, faithfully reproducing this diversity poses unique challenges. This paper presents an integrative review of simulation technologies for harvesting, pruning, and thinning, analyzing papers identified through dual-search strategy on Google Scholar and citation tracking. We synthesize methodologies along three core dimensions: (1) plant modeling, contrasting geometric techniques (e.g. NeRF, Gaussian Splatting) with physical models (MBD, FEM, Cosserat rods); (2) simulation platforms (e.g. Gazebo, Unity, Isaac Sim), evaluating trade-offs between real-time performance, visual fidelity, and physical accuracy; and (3) task-specific strategies distinguishing fruit detachment from branch pruning. Despite progress in synthetic data generation, our review identifies significant persisting gaps: specifically, the lack of standardized benchmarks for Sim-to-Real transfer and the difficulty of integrating flexible-body mechanics with real-time simulation. The paper concludes by outlining a roadmap toward hybrid plant modeling and unified evaluation frameworks, providing a foundation to accelerate the deployment of robust agricultural manipulators.
Hisashi Sugiura (2026) studied this question.