First Simulation Scenarios. First Results traffic impact of induced empty trips impact of delay on fleet performance. Network Information: ⢠25 km x 30 km.
Microsimulation of an Autonomous Taxi-System in Munich
Poster #5
Florian Dandl, Klaus Bogenberger – Munich University of the Federal Armed Forces
Autonomous Taxis (aTaxis) Big Players „Mercedes targets Silicon Valley rivals with robo-taxis by 2023“
„BMW and Intel to bring a fleet of selfdriving cars to the road by the end of 2017“
http://www.wired.co.uk/article/bmw-intel-driverless-tech-ces-2017
http://www.wired.co.uk/article/bmw-intel-driverless-tech-ces-2017
„Waymo taps Phoenix for a big public test of Google self-driving car tech“
„Self-driving Ubers: the world‘s first selfdriving Ubers are on the road in Steel City“
https://www.forbes.com/sites/alanohnsman/2017/04/25/waymotaps-phoenix-for-a-big-public-test-of-google-self-driving-cartech/#6a602eda6141
Research Questions and Framework
? •
•
„By 2030, individually owned ICE vehicles will still represent 40% of the vehicles in the U.S. vehicle fleet, but they will provide just 5% of passenger miles.“
inbound traffic
within-area trips O/D entries 10%
Traffic Volume
Travel Times
100%
modified private-vehicle O/D matrix (for all scenarios)
study area
possible requests (for aTaxis or for public transportation)
#(aTaxi requests) ------------------------------ 50% #(possible requests)
aTaxi Operation
Investigation of Traffic Impacts
other trips O/D entries
90%
transit traffic
vehicles might be tracked by users vehicle early, in-time or with small delay at customer location Incidents, creation of queues, traffic lights might cause longer delays ? re-routing ? re-assignment
First Results
original private-vehicle O/D matrix (base simulation)
outbound traffic
Traffic Situation
Travel Times: Estimated TT ≠ Driven TT
… „Transport-as-a-Service four to ten times cheaper per mile than buying a new car and two to four times cheaper than operating an existing vehicle in 2012.“
traffic impact of induced empty trips impact of delay on fleet performance
intra-city traffic
https://www.uber.com/cities/pittsburgh/self-driving-ubers/
Possibly Disruptive Technology
First Simulation Scenarios
60%
70%
80%
90%
100%
Operation Schematics
Customer 1 Origin
Destination
Implementation of Autonomous Taxi System in a Traffic Microsimulation
scenario
50
60
70
80
90
100
# aTaxi requests [x 1000]
20
24
28
32
36
40
fleet size
2000
2400
2800
3200
3600
4000
Customer 2 Origin
Destination
Selection of Model Assumptions •
Fleet Operator
Network Information: • • • • •
Autonomous ride of vehicle to customer Autonomous ride of vehicle with customer on board
3 request data sets per scenario generated by Poisson processes
Limitations and Planned Extensions •
ri = (request-time, start-location, destination)
Search for nearby vehicles and assignment of „optimal“ vehicle
Boundary of study area
25 km x 30 km 294 OD-centroids 2573 km total road length max. ~ 180k trips per hour 1238 aTaxi OD-locations (in residential and side roads)
• •
customer accept estimated waiting times of up to 10 minutes
•
each aTaxi serves only a single request at a time (no ride-pooling)
•
requests are unknown to the fleetoperation algorithm until request-time (on-demand-mobility)
•
vehicle, which will be available first at start-location of request, is assigned (can be in use at request-time!)
fleet size waiting times environmental impact empty vehicle mileage electrification ride-pooling financials …
Authors: • • • • • • •
Kockelman, Fagnant, Chen, Liu, … Frazolli, Spieser, Samaranayake, … Shaheen, Greenblatt, … Axhausen, Ciari, Boesch, Nagel, … Pavone, Zhang, Rossi, … Jung, Jayakrishnan, … …
Average Fleet Velocity [km / h]
Average Private Vehicle Delay [s / km]
50
7677 (19)
155 (1)
11.2 (0.1)
25.0 (0.1)
71.9 (1.6)
60
7678 (14)
185 (0)
10.7 (0.1)
25.0 (0.1)
70.3 (0.5)
70
7669 (4)
213 (0)
10.4 (0.1)
24.9 (0.1)
72.0 (0.3)
80
7682 (19)
245 (2)
10.1 (0.0)
24.6 (0.1)
72.4 (0.2)
90
7700 (10)
271 (1)
9.8 (0.1)
24.3 (0.1)
74.1 (1.6)
100
7715 (19)
303 (1)
9.6 (0.1)
24.2 (0.1)
75.1 (0.9)
Basis
7948 (8)
-
-
-
74.3 (0.4)
•
Investigation of More Realistic Traffic and Imperfect Knowledge on Fleet Performance • estimated waiting time: ETA computed by fleet operator based on estimated travel times
• real waiting time:
aTaxi demand is model input (high uncertainty, sensitivity analysis necessary)
• effective waiting time:
simplifications in current fleet-operation algorithms: • OD-routing by traffic simulator • does not consider or treat delays • global assignment not considered • vehicles stay put until request assignment
time from request until actual vehicle arrival
1) real waiting time for served requests 2) penalty time for unserved requests performance measure
…
Interface: Traffic Microsimulation & aTaxi Module
Previous Studies • • • • • • • •
Ratio of Empty Mileage [%]
long computation times, rule-of-thumb: cpu time [base scenario] + cpu time [link-level aTaxi simulation]
•
Research Questions:
aTaxi Mileage [1000 km]
average and standard deviation (in parantheses) of three replications with different seeds for private and aTaxi trip generation
http://www.rethinkX.com
simulation time period 05:00 – 11:00
Scenario
Private Vehicle Mileage [1000 km]
Simplified Travel Times: •
•
Perfect knowledge: the exact arrival time of vehicles is known to the operator
Aimsun API
Perfect operation: all vehicles drive exactly as long as predicted
aTaxi module
Traffic Situation
Travel Times
aTaxi Operation
aTaxi arrives at its destination
aTaxi enters next street section
update of link-level travel times for fleet computations
recording customer and vehicle statistics
update of vehicleavailability estimations
treatment of new customer requests
start of queued jobs
end of a time-step aTaxi starts driving in traffic simulation
trips to customers trips with customer on bord
recharging / refueling periodic relocations * * implementation in future framework
re-routing * delay-management of queued jobs * update of link-level travel times for fleet computations based on fleet-FCD *
NoMicroSim_75pc: time-independent link-level travel times NoMicroSim_TT: time-dependent link-level travel times (5 min updates)