ORCHARDBENCH
收藏资源简介:
ORCHARDBENCH是由昆士兰大学研究团队构建的一个基于物理引擎的GPU并行苹果园仿真基准数据集。该数据集通过随机L系统生成结构各异的苹果树模型,每棵树均由数百个柔性关节与独立果实刚体构成,并包含可调控密度的动态叶片层以模拟真实遮挡,数据规模支持在单个8GB笔记本电脑GPU上并行运行多个交互式仿真环境。数据集创建过程深度融合了植物学参数与物理动力学,所有机械参数均源自已发表的文献,实现了从语法生成骨架到完全铰接动力学体系的转化。该数据集旨在为农业机器人领域提供安全、可扩展的测试平台,重点解决果实采摘机器人研发中因实地实验成本高、不可重复及植物易损性带来的瓶颈问题,支持控制器验证、仿真到现实迁移学习、可控感知研究等多重应用场景。
ORCHARDBENCH is a physics-engine-based GPU-parallelized apple orchard simulation benchmark dataset developed by a research team at the University of Queensland. It generates apple tree models with diverse structures using stochastic L-systems, where each tree consists of hundreds of flexible joints and independent fruit rigid bodies, and is equipped with dynamically adjustable-density leaf layers to simulate realistic occlusion. The scale of the dataset supports parallel operation of multiple interactive simulation environments on a single 8GB laptop GPU. The dataset creation process deeply integrates botanical parameters and physical dynamics, with all mechanical parameters derived from published literature, realizing the transition from grammar-generated skeletons to fully articulated dynamic systems. This dataset aims to provide a safe and scalable test platform for the agricultural robotics domain, focusing on resolving bottleneck issues in fruit-picking robot development arising from high costs, irreproducibility, and plant vulnerability in field experiments, and supports multiple application scenarios including controller validation, simulation-to-reality transfer learning, and controllable perception research.




