Home

Welcome to the Simulator wiki!

Main logic

The simulator is based on a grid world and discrete time step. The grid world can be either hexagonal or quadrilateral. At the beginning of each time period, those activities happened:

Driver finished previous order and become available again.
Driver arrived the dispatched location based on allocation policy.
Generate orders.
Driver online offline status updates.
remove previous unserved orders.
----> state information collected and interact with agent <----
apply allocation policy delivered from agent.
assign orders to drivers and compute reward.

To be more specific, we can image each time step denotes a 10mins time interval. One episode in the simulator contains 144 time steps, which simulate one day's activities.

How to import your data into this simulator

1. Import real order data into the simulator

• mapped_matrix_int:
  The id of valid grid in the grid world. -100 indicates invalid grid.
• idle_driver_dist_time:
  A list of length 144. idle_driver_dist_time[0] = [10000, 200], the mean and std of total number of idle drivers for entire simulator at time step 0.
• idle_driver_location_mat:
  A 2d list of size 144 x number of grids.
  idle_driver_location_mat[0][grid_matrix_id] = 100 indicate mean of idle drivers at grid _grid_matrix_id_ at time 0.
• L_max The maximum time steps an order can across and the longest distance an order can across. If L_max = 9, the longest duration of order is 1.5 hours.
• M, N Size of matrix mapped_matrix_int
• n_side Indicate hexagonal (n_side = 6) or quadrilateral (n_side = 4) grid world.
• probability The probability of each order in real_orders being sampled.
• real_orders A matrix of size number of orders x 5.

real_orders[0] = [270.     9.   143.     2.    13.65]   
indicate start grid_id = 278, end grid_id = 9, order generated time = 143, order duration = 2, order price = 13.65.

• onoff_driver_location_mat A 3d array of size 144 x number of grid x 2.

onoff_driver_location_mat[0][1] = [-0.625       2.92350389], 
mean and std of changes of driver by online offline status updates at time step 0 and valid grid 1. env.target_node_ids[1] will relate this valid grid 1 to grid_id.

2. Import synthetic order data into the simulator

• order_num_dist:
  order_num_dist is list of length 144. Each element is a dictionary.

i.e. order_num_dist[0][22] = [mean, variance] 
indicates at time 0, the mean and std of number of orders on grid_matrix_id 22. 

• order_time_dist: A list of length L_max. The probability distribution of duration (order end time - order start time) of orders. If L_max = 9, the longest duration of order is 1.5 hours.

order_time_dist = [0.5, 0.4,  0.1] 
indicates there are 50% of orders last for 10 minutes, 40% last for 20 minutes and 10% last for 30 minutes (3 time steps). 

• order_price_dist: A 2d list of size L_max x 2

order_price_dist[0] = [10, 2] 
indicate the mean and std of orders' price for those orders last 10 minutes.

NOTICE: grid_matrix_id is the index of grid in this matrix mapped_matrix_int . grid_id is the id of valid grid set by yourself. They can be same.