2
$\begingroup$

A cleaner, denoted by $P$, aims to sweep $n\ge 1$ leaves that appear one by one in a courtyard modeled by a compact set $D\subset \mathbb R^2$. Denote by $x_0$ the initial position of $P$ and by $v>0$ his moving speed. Let $\mu$ be a probability measure on $D$ and $d$ be some distance function on $D$. Assume that the leaves appear independently with their position distributed according to $\mu$.

More precisely, for each $1\le i\le n$, suppose $P$ has just swept the $(i-1)^\text{th}$ leaf and his position is denoted by $X_{i-1}$. Let $t>0$ be the waiting time for the $i^\text{th}$ leaf since the $(i-1)^\text{st}$ leaf has been swept. $P$ may move to any position $y\in B(X_{i-1}, vt)\mathrel{:=}\{y\in D: d(X_{i-1},y)\le vt\}$. Let $y_i \in B(X_{i-1}, vt)$ be the position to which $P$ moves from $X_{i-1}$ during the waiting time. As the $i^\text{th}$ leaf appears at $X_i$, $P$ moves further to $X_i$. Set $X_0\mathrel{:=}x_0$. How to minimize the average time $$\rlap{ \mathbb E\left[\sum_{i=1}^n \left(t+\frac{d(y_i, X_{i})}{v}\right)\right]} \hphantom{\mathbb E\left[\sum_{i=1}^n \left(d(X_{i-1},y_i)+d(y_i, X_{i})\right)\right]?}$$ over all $y_1,\ldots, y_n$ s.t. $y_i\in B(X_{i-1}, vt)$ for $1\le i\le n$? Similarly, how to minimize the average distance $$\mathbb E\left[\sum_{i=1}^n \left(d(X_{i-1},y_i)+d(y_i, X_{i})\right)\right]?$$

Personal thought : Define $f:D\to\mathbb R_+$ by

$$f(x):=\min_{y\in B(x,vt)}\mathbb E\left[t+\frac{d(y,X)}{v}\right],$$

where $X\sim \mu$ is a random variable. Introduce recursively $f_n:=f$ and for $0\le k\le n-1$ :

$$f_{k}(x):=\min_{y\in B(x,vt)} \mathbb E\left[t+\frac{d(y,X)}{v}+f_{k+1}(X)\right] = f(x) + \mathbb E[f_{k+1}(X)].$$

Hence, the minimization problem is identified by $f_0(x_0)$. Then my question can be reformulated : How to solve efficiently this optimization problem?

$\endgroup$
5
  • 1
    $\begingroup$ This may be relevant: Arkin, Esther M., Sándor P. Fekete, and Joseph SB Mitchell. "Approximation algorithms for lawn mowing and milling." Computational Geometry 17, no. 1-2 (2000): 25-50. DOI. $\endgroup$ Commented Jun 6, 2021 at 0:06
  • $\begingroup$ @JosephO'Rourke Thanks for the reference. Computational geometry is a completely new domain for me. I will read this paper $\endgroup$
    – user128095
    Commented Jun 6, 2021 at 0:16
  • $\begingroup$ I would recommend $P$ wait until all the leaves have fallen... All jokes aside, this seems difficult to answer because $\mu$ could be quite complicated. $\endgroup$
    – Leo Moos
    Commented Jun 6, 2021 at 6:47
  • $\begingroup$ @LeoMoos Indeed this is what I thought... Actually this is a game of my son, where leaf $i+1$ appears only after leaf $i$ is swept. In this game $D=[0,1]\times [0,1]$. So I think $mu$ can be chosen as the uniform distribution on $D$. $\endgroup$
    – user128095
    Commented Jun 6, 2021 at 10:11
  • $\begingroup$ It seems to me that for convex $D$, with uniform $\mu$, your son should walk back to the center of mass. (When $D$ is not convex, the story seems a bit more complicated.) $\endgroup$
    – Leo Moos
    Commented Jun 6, 2021 at 16:53

0

You must log in to answer this question.