**Input:** The graph \(G\), the parameter \(K\), and a starting solution \(s\) (could be empty) |

**Output:** A solution for the current time step |

1: Compute \(KG;\) |

2: **while** time is available **do** |

3: **while** exploration time is available **do** |

4: For each customer \(c\), generate uniformly a pick-up time \(t_{c} \in I_{c}\) or \(t_{c} \in I_{c}^{s};\) |

5: Solve maxFlow on \(KG\) with \(t_{c}\) pick-up times of customer; |

6: Add the arc profits of the solution to features \(X\) and \(\left( t_{c} \right)\) to \(y;\) |

7: **end while** |

8: Train the RNN model \(m\) using \(X\) and \(y;\) |

9: Generate times by \(\left. \ \mathbf{t} = m\left( \left\lbrack R_{e_{1}},\ldots,R_{|E|} \right) \right\rbrack^{'} \right);\) |

10: Solve maxFlow with times \(\mathbf{t};\) |

11: Update the solution \(s;\) |

12: **end while** |