• Taxi

Introduction#

Since version 1.5.0 SUMO supports simulation of demand responsive transport (DRT) via the taxi device. This allows a fleet of taxis to service customer requests based on a configurable dispatch algorithm.

Equipping Vehicles#

A vehicle can be equipped with an Taxi device to make it part of the taxi fleet. To attach a Taxi device to a vehicle, the standard device-equipment procedures can be applied using <device name>=taxi.

For instance, a single vehicle can configured as taxi as in the following minimal example

    <vehicle id="v0" route="route0" depart="0" line="taxi">
        <param key="has.taxi.device" value="true"/>
    </vehicle>

Taxi Requests#

Direct Ride Hailing#

A person can be defined as taxi customer with the following definition:

    <person id="p0" depart="0.00">
        <ride from="B2C2" to="A0B0" lines="taxi"/>
    </person>

Intermodal Routing#

A person can also use a taxi by including it as a personTrip mode:

    <person id="p0" depart="0.00">
        <personTrip from="B2C2" to="A0B0" modes="taxi"/>
    </person>

Whenever a person enters a taxi during the intermodal route search, a time penalty is applied to account for the expected time loss from waiting for the taxi and embarking. The default value is set 300s and can be configure with option --persontrip.taxi.waiting-time. This prevents rapid switching between travel modes.

Groups of Persons#

Multiple persons can travel together as a group using attribute group (if the taxi has sufficient capacity):

    <person id="p0" depart="0.00">
        <ride from="B2C2" to="A0B0" lines="taxi" group="g0"/>
    </person>
    <person id="p1" depart="0.00">
        <ride from="B2C2" to="A0B0" lines="taxi" group="g0"/>
    </person>

Prebooking#

Direct ride hailing can be regarded as a spontaneous booking. If the taxi is to be requested before the person is at the pick-up location, prebooking is necessary. This is used by specifying an earliestPickupTime and a reservationTime in the ride. The reservationTime specifies the time at which the reservation is created and made available to the dispatch. It may also be less than the person's depart. The earliestPickupTime specifies the time at which the person is ready to be picked up. Depending on the algorithm used, this may allow a better dispatch to be made. Especially in combination with a custom dispatch algorithm via TraCi, the requirements of different use cases can be addressed.

Application:#

• The file with the persons must be loaded as an additional file.
• At least the earliestPickupTime must be specified. The reservationTime has the start time of the simulation as the default value.
• It is necessary to define the attribute from in the ride.
• If there was previously a walk in which the arrivalPos attribute was defined, then the fromPos attribute must also be defined in the ride with the same value.

Example:#

    <person id="p0" depart="100.00">
        <ride from="B2C2" to="A0B0" lines="taxi">
            <param key="earliestPickupTime" value="100.00"/>
            <param key="reservationTime" value="50.00"/>
        </ride>
    </person>
    <person id="p1" depart="0.00">
        <walk from="B2C2" to="B1C1"/>
        <ride from="B1C1" to="A1B1" lines="taxi">
            <param key="earliestPickupTime" value="30.00"/>
        </ride>
    </person>

Behavior:#

• If a person arrives at the departure point earlier than the earliestPickupTime, the person boards as soon as the taxi arrives.
• If the taxi is on time and the person does not arrive at the departure point within 3 minutes of the earliestPickupTime, the reservation is canceled and the taxi serves the next reservation.
• If the taxi arrives later and the person does not arrive at the departure point within 3 minutes of the taxi's arrival time, the reservation is canceled.
• If the greedy or greedyShared scheduling algorithms are used, the reservations are processed in the order of the earliestPickupTime. A combination of spontaneous bookings and prebookings can lead to side effects. It is recommended to use a custom dispo algorithm via TraCi.

Notes:#

• Prebookings for merged reservations and group reservations are not yet supported.
• The current status can be tracked in Ticket 11429

Multiple Taxi Fleets#

By default, there is only a single taxi fleet using line attribute 'taxi' and taxi customers use attribute lines="taxi" for their rides. It is permitted to define the line attribute for taxi with the prefix 'taxi:' and an arbitrary suffix (i.e. "taxi:fleetA"). Likewise, it is permitted to define the lines attribute for rides with the prefix 'taxi:' and a suffix. When this is done, the following rules are applied when assigning taxis to customers:

• a taxi with line 'taxi:X' may only pick up customers with matching ride attribute lines="taxi:X" (for any value of X)
• a customer with lines="taxi" may use any taxi regardless of taxi fleet suffix
• a taxi with line 'taxi' may be pick up any customer regardless of the ride fleet suffix

Dispatch Algorithms#

The dispatch algorithm assigns taxis to waiting customers. The algorithm is selected using option --device.taxi.dispatch-algorithm ALGONAME. The following algorithms are available

• greedy: Assigns taxis to customers in the order of their reservations. The closest taxi (in terms of travel time) is assigned. If the reservation date is too far in the future, the customer is postponed.

• greedyClosest: For each available taxi, the closest customer (in terms of travel time) is assigned. If the reservation date is too far in the future, the customer is postponed.

• greedyShared: like 'greedy' but tries to pick up another passenger while delivering the first passenger to its destination. Parameters absLossThreshold and relLossThreshold to configure acceptable detours can be supplied using --device.taxi.dispatch-algorithm.params KEY1:VALUE1[,KEY2:VALUE].

• routeExtension: like greedy but can pick up any passenger along the route and also extend the original route (within personCapacity limit).

• traci: Dispatch is deferred to traci control. The algorithm only keeps track of pending reservations

Taxi Behavior#

By default, taxis will remain in the simulation until all persons have left. To make them leave the simulation at an earlier time, the end time can be defined using a generic parameter in their vType or vehicle-definition:

    <vType id="taxi" vClass="taxi">
        <param key="has.taxi.device" value="true"/>
        <param key="device.taxi.end" value="3600"/>
    </vType>

Idle Behavior#

By default, vehicles will leave the simulation after reaching the end of their final edge. To avoid this, taxis have an idling behavior configurable with option --device.taxi.idle-algorithm:

• "stop" (default): stop at the current location (off-road) after delivering the last customer of the current service request.
• "randomCircling": continue driving to random edges until the next request is received. (caution: taxi might get stuck in a cul-de-sac if the network has those)
• "taxistand": drive to a taxi stand and wait there for the next customer / dispatch. Defining the set of taxi stands and the strategy of choosing among them is described below

Defining taxi stands#

When using idle-algorithm taxistand, the following inputs must be provided:

• each taxi stand must be defined as a parkingArea
• the list of parkingAreas that may be used for a particular taxi or taxi fleet must be defined as a <rerouter>-element according to the description for parking search simulation.
• the taxi must define the parameter device.taxi.stands-rerouter either as a child element of the <vehicle> or its <vType> and declare the rerouter id.

The strategy for choosing among the alternative taxi stands follows the description for parking search simulation (i.e. with respect to prior knowledge of remaining capacity).

Example declarations for the rerouter and the taxi vType that references it:

<rerouter id="rr0" edges="B0C0 E2D2" vTypes="taxi">
        <interval begin="0" end="1:0:0:0">
            <parkingAreaReroute id="pa_0"/>
            <parkingAreaReroute id="pa_1"/>
        </interval>
</rerouter>

<vType id="taxi" vClass="taxi">
        <param key="has.taxi.device" value="true"/>
        <param key="device.taxi.stands-rerouter" value="rr0"/>
</vType>

Customer Stops#

Taxis will stop to pick-up and drop-off customers. The 'actType' attribute of a stop indicates the purpose ('pickup' / 'dropOff') as well as the ids of the customers and their reservation id. Stop attributes can be configured using generic parameters <vType> or <vehicle> definition of the taxi:

    <vType id="taxi" vClass="taxi">
        <param key="has.taxi.device" value="true"/>
        <param key="device.taxi.pickUpDuration" value="0"/>
        <param key="device.taxi.dropOffDuration" value="60"/>
        <param key="device.taxi.parking" value="false"/>
    </vType>

• duration for pick-up stop can be configured with vType/vehicle param "device.taxi.pickupDuration" (default "0")
• duration for drop-off stop can be configured with vType/vehicle param "device.taxi.dropOffDuration" (default "60")

By default, vehicle stops will have attribute parking="true" which means that the taxi will not block a driving lane. This can be changed by setting param "device.taxi.parking" to "false".

TraCI#

To couple an external dispatch algorithm to SUMO, the following TraCI functions are provided:

• traci.person.getTaxiReservations(reservationState)
• traci.vehicle.getTaxiFleet(taxiState)
• traci.vehicle.dispatchTaxi(vehID, reservations)

This set of API calls can be used to simplify writing custom dispatch algorithms by letting sumo:

• manage existing reservations
• manage the taxi fleet
• dispatch a taxi to service one or more reservations by giving a list of reservation ids (vehicle routing and stopping is then automatic).

getTaxiReservations#

Returns a list of of Reservation objects that have the following attributes

• id
• persons
• group
• state
• fromEdge
• toEdge
• arrivalPos
• departPos
• depart
• reservationTime
• state (positive value, see below)

When calling traci.person.getTaxiReservations(reservationState) the following arguments for reservationState are supported:

• 0: return all reservations regardless of state
• 1: return only new reservations
• 2: return reservations already retrieved
• 4: return reservations that have been assigned to a taxi
• 8: return reservations that have been picked up

Combinations of these values are also supported. For example sending a value of 3 (= 1 + 2) will return all reservations of both states 1 and 2.

getTaxiFleet#

A taxi can be in any of the following states:

• 0 (empty) : taxi is idle
• 1 (pickup): taxi is en-route to pick up a customer
• 2 (occupied): taxi has customer on board and is driving to drop-off
• 3 (pickup + occupied): taxi has customer on board but will pick up more customers

when calling traci.vehicle.getTaxiFleet(taxiState) the following arguments for taxiState are supported:

• -1: (return all taxis regardless of state)
• 0: return only empty taxis
• 1: return taxis in state 1 and 3
• 2: return taxis in state 2 and 3
• 3: return taxis in state 3

dispatchTaxi#

If a taxi is empty, the following dispatch calls are supported

• dispatchTaxi(vehID, [reservationID]): pickup and drop-off persons belonging to the given reservation ID
• If more than one reservation ID is given, each individual reservation ID must occur exactly twice in the list for complete pickup and drop-off. The first occurrence of an ID denotes pick-up and the second occurrence denotes drop-off.

Example 1: dispatchTaxi(vehID, [a]) means: pick up and drop off a.

Example 2: dispatchTaxi(vehID, [a, a, b, c, b, c]) means: pick up and drop off a, then pick up b and c and then drop off b and c.

If a taxi is not in state empty the following re-dispatch calls are supported

• new reservations have no overlap with previous reservation: append new reservations to the previous reservations
• new reservations include all previous unique reservation ids exactly twice: reset current route and stops and treat as complete new dispatch. If one of the persons of the earlier reservation is already picked up, ignore the first occurrence of the reservation in the reservation list
• new reservations mentions include all previous unique reservation ids once or twice, all customers that are mentioned once are already picked up: reset current route and stops, use the single-occurrence ids as as drop-of

Outputs#

The Taxi device generates output within a tripinfo-output file in the following form:

    <tripinfo id="trip_0" ... >
        <taxi customers="5" occupiedDistance="6748.77" occupiedTime="595.00"/>
    </tripinfo>

Parameter Retrieval#

The following parameters can be retrieved via traci.vehicle.getParameter and written via --fcd-output.params. It is also possible to color vehicles in SUMO-GUI 'by param (numerical)' by setting these keys.

• device.taxi.state: returns integer value (see #gettaxifleet)
• device.taxi.customers: total number of customers served
• device.taxi.occupiedDistance: total distance driven in m with customer on board
• device.taxi.occupiedTime: total time driven in s with customers on board
• device.taxi.currentCustomers: space-separated list of persons that are to be picked up or already on board