3万人草坪路径实验揭示人类直觉算法
推荐指数 33.0 NO. 022 · 2026.08.05
发布2026/08/04Score55Comments40
为什么值得看
一项在线实验让30954人模拟 mowing lawn 路径规划，52%接近最优解，16%完美完成。人类在无意识中解决的覆盖路径问题，对机器人导航和自动驾驶的启发式算法设计有直接参考价值。
编辑判断
这个实验的巧妙之处在于把计算几何里的经典 NP-hard 问题（覆盖路径规划）转化成了人人可参与的直觉测试。传统上机器人用 boustrophedon 或螺旋算法，但人类显然在用某种空间分块的贪心策略，且效果惊人地好。
值得深挖的是那 16% 完美解的人群特征：他们是否共享某种空间认知模式？这种直觉能否被蒸馏成新的近似算法？做自动驾驶路径规划或仓储机器人调度的团队，可以把这个实验的数据集要过来做对比分析，可能比现有启发式规则更鲁棒。
社区反馈
意见分歧 48 条评论
核心争论：抽象路径优化 vs 真实场景多目标优化：研究简化了实际 mowing 的复杂约束
相关内容
Though I solved it in 49 moves (yay me), I realized when thinking about efficiency, it didn't take into consideration two important things: 1) all of the turns I was making in order to be most efficient, whereas overlapping a few times would have probably taken less time, and 2) I spent a lot of thi
This is a very different problem than actual efficiency when mowing a lawn or vacuuming a rub. Turning takes more work and time, and you miss a part of the 'square' when you are turning (the arc of the curve is not a straight line). In addition, vacuums especially have an area at the edge of the mac
I was thinking the same thing. When actually mowing a lawn (or similarly, vacuuming a carpet/rug) I tend to optimize for either long lines or rectangular sections where I can easily mow a straight pattern.