论文arXiv 2611.07293今天

遗传算法优化果蔬采摘序：动态航点策略提升接近成功率至70.7%

针对高密度非结构化环境，提出结合RGB-D定位与遗传算法的机器人采摘排序方法。田间测试显示，动态航点连续执行策略较静态策略将目标接近成功率从65.3%提升至70.7%，碰撞事件减少27%，验证了全局规划与局部感知协同的有效性。

AI 整理

感知与导航
具身模型

本文由 AI 根据公开内容整理，可能有误，请以原文为准：An Ordering Optimization for Fruit-Vegetable in High-Density Environment based on Genetic Algorithm（arXiv 2611.07293）

核心内容

本文提出一种面向高密度非结构化作物环境的机器人采摘方法，融合全局RGB-D相机三维定位、末端近距离观测及基于遗传算法（GA）的采摘序列优化。研究在实验室及4至6果田间场景完成评估，通过参数敏感性分析对比了静态与动态航点对运动效率的影响，证实连续执行策略可显著减少无效回零运动，为复杂农业环境下的自动化作业提供量化依据。

系统采用全局RGB-D获取果实坐标供GA排序，末端相机负责执行期近距感知。核心创新在于动态航点机制：在不改变GA既定序列前提下，根据剩余果实分布实时调整中间路径点。田间实测表明，该策略使目标接近成功率达70.7%（静态为65.3%），碰撞次数由55次降至40次，证明了自适应路径规划在降低物理干涉风险方面的工程价值。

关键看点

动态航点机制：不改变GA序列，依据剩余果实分布实时更新中间路径点，消除固定航点导致的冗余运动与潜在碰撞风险
双级视觉感知：全局RGB-D相机提供3D坐标用于排序规划，末端相机在执行阶段进行近距离观测校正，适配非结构化环境
田间实测数据：动态航点连续策略实现70.7%目标接近成功率，较静态策略提升5.4个百分点，碰撞事件减少27%
连续执行优化：摒弃传统单果采摘后返回Home点模式，通过 waypoints 定义中间运动，大幅压缩无效空行程时间

工程与产业观察

该研究揭示了在非结构化农业场景中，单纯依赖全局最优序列不足以解决物理执行问题。动态航点策略将离散的组合优化与连续的几何约束解耦，为具身智能在密集障碍物环境下的任务规划提供了“全局排序+局部自适应”的工程范式，对提升采摘机器人实际作业鲁棒性具有直接参考价值。

英文摘要 (Abstract)

This paper presents a robotic harvesting method that combines RGB-D fruit localization, GA-based harvest ordering, and waypoint-based robot motion in dense and unstructured crop environments. The global RGB-D camera provides 3D fruit positions for ordering, while the end-effector-mounted camera provides close-range observation during execution. The GA determines the harvesting sequence, and static or dynamic waypoints define the intermediate robot motion. The method was evaluated in a controlled laboratory testbed and in four-, five-, and six-fruit field conditions. GA solutions were compared with exhaustive enumeration, and both waypoint offsets were examined by parameter sensitivity analysis. Continuous execution reduced unnecessary Home-return motion, while the dynamic waypoint changed with the remaining fruit distribution without changing the GA sequence. In the field evaluation, the dynamic-waypoint continuous strategy produced 40 collision events and a 70.7% target-approach success rate, compared with 55 collision events and 65.3% for the static-waypoint continuous strategy.

作者：Jaehwi Seol, Jeonghyeon Pak, Hyoung Il Son