Cost model parameters

AI-generated Key Takeaways

Route Optimization's cost model, defined by various cost parameters, guides the optimization process by enabling trade-offs between factors like time, distance, and shipment completion.
Cost parameters, expressed in dimensionless units, are assigned to Vehicle and Shipment properties, allowing users to prioritize route characteristics and shipment importance.
Shipment penalty costs influence whether a shipment is included in a route; if the cost to complete a shipment exceeds its penalty cost, it may be skipped.
The OptimizeToursResponse message provides cost breakdowns, including metrics.costs and routeCosts, offering insights into the cost efficiency of the generated routes.
Soft constraints, such as LoadLimit and TimeWindow, introduce cost penalties when violated, allowing for flexibility while still encouraging adherence to limits.

European Economic Area (EEA) developers

The OptimizeToursRequest message (REST, gRPC) contains a number of properties related to costs. Together, these cost parameters represent the request's cost model. The cost model captures many of the request's high-level optimization objectives, such as:

Prioritizing faster Vehicle routes over shorter routes or the other way around
Deciding whether the cost to deliver a Shipment is worth the value of the Shipment's completion
Performing pickups and deliveries within time windows only when doing so is cost effective

See an example request with costs

{
  "model": {
    "globalStartTime": "2023-01-13T16:00:00-08:00",
    "globalEndTime": "2023-01-14T16:00:00-08:00",
    "shipments": [
      {
        "deliveries": [
          {
            "arrivalLocation": {
              "latitude": 37.789456,
              "longitude": -122.390192
            },
            "duration": "250s"
          }
        ],
        "pickups": [
          {
            "arrivalLocation": {
              "latitude": 37.794465,
              "longitude": -122.394839
            },
            "duration": "150s"
          }
        ],
        "penaltyCost": 100.0
      },
      {
        "deliveries": [
          {
            "arrivalLocation": {
              "latitude": 37.789116,
              "longitude": -122.395080
            },
            "duration": "250s"
          }
        ],
        "pickups": [
          {
            "arrivalLocation": {
              "latitude": 37.794465,
              "longitude": -122.394839
            },
            "duration": "150s"
          }
        ],
        "penaltyCost": 5.0
      },
      {
        "deliveries": [
          {
            "arrivalLocation": {
              "latitude": 37.795242,
              "longitude": -122.399347
            },
            "duration": "250s"
          }
        ],
        "pickups": [
          {
            "arrivalLocation": {
              "latitude": 37.794465,
              "longitude": -122.394839
            },
            "duration": "150s"
          }
        ],
        "penaltyCost": 50.0
      }
    ],
    "vehicles": [
      {
        "endLocation": {
          "latitude": 37.794465,
          "longitude": -122.394839
        },
        "startLocation": {
          "latitude": 37.794465,
          "longitude": -122.394839
        },
        "costPerHour": 40.0,
        "costPerKilometer": 10.0
      }
    ]
  }
}

`Vehicle` cost properties

The Vehicle message (REST, gRPC) has several cost properties:

Vehicle.cost_per_hour: represents the cost of operating a vehicle per hour inclusive of transit, wait, visit, and break times..
Vehicle.cost_per_kilometer: represents the cost per kilometer traveled by the vehicle.
Vehicle.cost_per_traveled_hour: represents the cost of operating a vehicle only while in transit, excluding wait, visit, and break times.

These cost parameters allow the optimizer to make time-versus-traveled-distance tradeoffs. The costs incurred by the optimized route appear in the response message as metrics.costs:

As costPerHour increases, the optimizer attempts to find faster routes that may not be the shortest routes. In this example the fastest route happens to be the shortest, so changes to the cost parameters have little effect.

`Shipment` cost properties

The Shipment message (REST, gRPC) also has several cost parameters:

Shipment.penalty_cost represents the cost incurred by skipping the shipment. Not setting a shipment's penalty_cost parameter makes the shipment mandatory, which means the shipment will only be skipped if it cannot be completed given specified constraints.
Shipment.VisitRequest.cost represents the cost of a specific pickup or delivery, used primarily to enable cost tradeoffs between multiple pickup or delivery options for a single shipment.

Shipment cost parameters use the same dimensionless units as Vehicle cost parameters. Cost incurred complete a Shipment exceeds its penalty cost, the Shipment is not included on any Vehicle's route and instead appears in the skipped_shipments list in the response message.

`ShipmentModel` cost properties

The ShipmentModel message (REST, gRPC) includes a single cost property, globalDurationCostPerHour. This cost is incurred based on the total time required for all vehicles to complete their ShipmentRoutes. Increasing globalDurationCostPerHour prioritizes earlier completion of all shipments.

Route Optimization response cost properties

The OptimizeToursResponse message (REST, gRPC) has cost properties that represent the costs incurred in the process of completing ShipmentRoutes. The metrics.costs and metrics.totalCost properties represent the number of cost units incurred across all routes in the response. Each routes entry has routeCosts and routeTotalCosts properties that represent costs for that specific route.

See a response to the example request with costs

{
  "routes": [
    {
      "vehicleStartTime": "2023-01-14T00:00:00Z",
      "vehicleEndTime": "2023-01-14T00:28:22Z",
      "visits": [
        {
          "isPickup": true,
          "startTime": "2023-01-14T00:00:00Z",
          "detour": "0s"
        },
        {
          "shipmentIndex": 2,
          "isPickup": true,
          "startTime": "2023-01-14T00:02:30Z",
          "detour": "150s"
        },
        {
          "startTime": "2023-01-14T00:08:55Z",
          "detour": "150s"
        },
        {
          "shipmentIndex": 2,
          "startTime": "2023-01-14T00:21:21Z",
          "detour": "572s"
        }
      ],
      "transitions": [
        {
          "travelDuration": "0s",
          "waitDuration": "0s",
          "totalDuration": "0s",
          "startTime": "2023-01-14T00:00:00Z"
        },
        {
          "travelDuration": "0s",
          "waitDuration": "0s",
          "totalDuration": "0s",
          "startTime": "2023-01-14T00:02:30Z"
        },
        {
          "travelDuration": "235s",
          "travelDistanceMeters": 795,
          "waitDuration": "0s",
          "totalDuration": "235s",
          "startTime": "2023-01-14T00:05:00Z"
        },
        {
          "travelDuration": "496s",
          "travelDistanceMeters": 1893,
          "waitDuration": "0s",
          "totalDuration": "496s",
          "startTime": "2023-01-14T00:13:05Z"
        },
        {
          "travelDuration": "171s",
          "travelDistanceMeters": 665,
          "waitDuration": "0s",
          "totalDuration": "171s",
          "startTime": "2023-01-14T00:25:31Z"
        }
      ],
      "metrics": {
        "performedShipmentCount": 2,
        "travelDuration": "902s",
        "waitDuration": "0s",
        "delayDuration": "0s",
        "breakDuration": "0s",
        "visitDuration": "800s",
        "totalDuration": "1702s",
        "travelDistanceMeters": 3353
      },
      "routeCosts": {
        "model.vehicles.cost_per_kilometer": 33.53,
        "model.vehicles.cost_per_hour": 18.911111111111111
      },
      "routeTotalCost": 52.441111111111113
    }
  ],
  "skippedShipments": [
    {
      "index": 1
    }
  ],
  "metrics": {
    "aggregatedRouteMetrics": {
      "performedShipmentCount": 2,
      "travelDuration": "902s",
      "waitDuration": "0s",
      "delayDuration": "0s",
      "breakDuration": "0s",
      "visitDuration": "800s",
      "totalDuration": "1702s",
      "travelDistanceMeters": 3353
    },
    "usedVehicleCount": 1,
    "earliestVehicleStartTime": "2023-01-14T00:00:00Z",
    "latestVehicleEndTime": "2023-01-14T00:28:22Z",
    "totalCost": 57.441111111111113,
    "costs": {
      "model.vehicles.cost_per_kilometer": 33.53,
      "model.vehicles.cost_per_hour": 18.911111111111111,
      "model.shipments.penalty_cost": 5
    }
  }
}

In the example response, the top-level metrics.costs are:

{
  "metrics": {
    ...
    "costs": {
      "model.vehicles.cost_per_hour": 18.911111111111111,
      "model.vehicles.cost_per_kilometer": 33.53,
      "model.shipments.penalty_cost": 5
    }
  }
}

The model.shipments.penalty_cost value represents the cost incurred due to skipped shipments. The skippedShipments property lists which shipments were skipped.

In this example, only model.shipments[1] in the example request is skipped. model.shipments[1] has a penalty cost of 5 units, which matches the total model.shipments.penalty_cost key in the example response. The shipment's low penaltyCost compared to the Vehicle's 40.0 costPerHour and 10.0 costPerKilometer make it more cost-effective to skip the shipment than to complete it.

Advanced topic: costs and soft constraints

Several OptimizeToursRequest message (REST, gRPC) properties represent soft constraints, which are constraints that incur a cost when they cannot be satisfied.

For example, vehicle LoadLimit (REST, gRPC) constraints have softMaxLoad and costPerUnitAboveSoftMax properties. Together, these incur a cost proportional to the load units that exceed softMaxLoad, allowing the limit to be exceeded only if doing so makes sense from a cost standpoint.

Similarly, TimeWindow constraints (REST, gRPC) has soft_start_time and soft_end_time properties, with corresponding cost_per_hour_before_soft_start_time and cost_per_hour_after_soft_end_time that are incurred based on how early or late the constrained event occurs with respect to the TimeWindow.

As with all cost model parameters, soft constraint costs are expressed in the same dimensionless units as other cost parameters.

LoadLimit constraints are addressed in detail in Load Demands and Limits. TimeWindow constraints are addressed in detail in Pickup and Delivery Time Window Constraints.