Applications: Last mile
subsidies, mobility hubs, and
technology deployment
Joseph Chow
Institute Associate Professor
June 26, 2024
Outline
• Illustrations of MaaS scenario analysis with tool
• Designing subscription plans
• TRB research brainstorming session:
• Mobility hub design as special case
• Transit fixed route extension vs last mile on-demand feeder
• Simulation-based algorithm evaluation in MaaS setting
• Parameter learning/updating with deployment data
• Alternative (path-based, non-additive) pricing schemes
• Congestion pricing scenario analysis
• Equity analysis: sufficiency maximization
• Adapting design to freight deliveries
2
Sioux Falls scenario
illustrations
3
Overview
• Sioux Falls network cannot be solved using commercial solver, with 30
OD pairs, 82 nodes, and 748 links in the expanded network
• Policy sensitivity of model tested on different scenarios:
• Impact when platform subsidizes either all MOD users vs one operator
• Identifying equity impact when platform’s users are in fact heterogeneous
4
Sioux Falls Example
30 OD pairs:
OD ID
1
Origin
2
Destination
1
Demand
100
OD ID
16
Origin
1
Destination
18
Demand
100
2 12 1 200 17 2 18 0
3 18 1 100 18 12 18 200
4 13 1 500 19 13 18 100
5 20 1 300 20 20 18 400
6 1 2 100 21 1 13 500
7 12 2 100 22 2 13 300
8 18 2 0 23 12 13 1300
9 13 2 300 24 18 13 100
10 20 2 100 25 20 13 600
11 1 12 200 26 1 20 300
12 2 12 100 27 2 20 100
13 18 12 200 28 12 20 400
14 13 12 1300 29 18 20 400
15 20 12 500 30 13 20 600
MOD access cost of access link :
, where . Expanded
MOD operator cost of MOD link Network:
,
82 nodes
Utility: for all OD pairs.
Alternative MOD fleet sizes: 1 and 2
748 links
5
Base Case Results
Exact solution found:
Buyer Opt Seller Opt
1 $4800 $15,300
2 \ \
Operato 3 \ \
r 4 \ \
Revenue 5 (MOD) \ \
6 (MOD) \ \
7 (MOD) \ \
User Payoff $85,200 $74,700
Total system cost $108,800
Unserved demand 1200
Run time 4 hrs 1 min
6
Last mile subsidy:
What if MOD operator
5 is subsidized 50%?
Subsidized MaaS platform user
equilibrium:
Buyer Opt Seller Opt
1 $4800 $11,099 (27%)
2 \ \
3 \ \
Operator
4 \ \
Revenue
5 (MOD) $66.25 $94.71
6 (MOD) $5.17 $5.17
7 (MOD) \ \
User Payoff $85,227 () $78,900 ()
Stabilizing subsidy $4.09
Total system cost $108,781 Subsidy for
Unserved demand 1188 () Operator 5
incentivizes
Operator 6
Run time: 10 hrs 23 min
market entry
7
Multimodal
paths
incentivized
by subsidy to
Operator 5
8
What if the users Demand of all OD pairs are split into 2 groups:
• Group 1 (1/2 demand): trip utility = 24 (higher income)
are heterogeneous? • Group 2 (1/2 demand): trip utility = 16 (lower income)
Such that they average to the base case (utility = 20 for all users).
Subsidized MaaS platform user equilibrium found:
Buyer Opt Seller Opt
1 $4800 $21,750 ()
2 \ \
3 \ \
Operator
4 \ \
Revenue
5 (MOD) $39.52 $56.08
6 (MOD) $5.11 $5.11
7 (MOD) \ \
User Payoff $86,018 () $69,051 ()
Stabilizing subsidy $3.94
Total system cost $107,990 ()
2243.22 ():
Unserved demand Equity issue
HI: 594, LI: 1650
among different
income groups!
Run time: 12 hrs 33 min
MOD service used only by Group 1 9
Summary
• Platform equilibrium achieved with or without subsidy
• Model sensitive to changes in operating costs, travel costs, external
utilities, network structure
• Algorithm obtains exact or bounded solutions
10
Analyzing subscription
plans
11
How to set a subscription price?
What is impact on the market?
• Unlimited ride plans (daily/weekly/monthly/annual/etc.)
• Single ticket (regardless of route/mode) by…
• OD pair
• By population segment (age/student/income) regardless of OD or
mode/route
• By OD pair regardless of population segment or mode/route
12
Ex: consider OD-based fare
Instead of link-based fares, users with subscriptions pay at OD level
• Matching problem is the same
• Stability conditions modified to use instead of
D D
𝑝 𝑂𝐷
𝑝1 𝐷
1 1
O O
𝑝𝑂 1
Pay per ride Fee charged to OD group users
13
How would implementing an
OD-based fare compare to pay
per ride?
OD based subscription can significantly increase
seller-optimal revenues
14
How does OD-based subscription respond
when Operator 4 operating cost reduced 60%?
15
Brainstorming (TRB)
research topics
Mobility hub design as special case
Transit fixed route extension vs last mile on-demand feeder
Simulation-based algorithm evaluation in MaaS setting
Parameter learning/updating with deployment data
Alternative (path-based, non-additive) pricing schemes
Congestion pricing scenario analysis
Adapting design to freight deliveries
16
How to locate mobility hubs, allocate
operators, and size parking and subsidies?
• Pre-req: data on existing hub investment(s) and zonal demand to
calibrate model,
• Contribution: method to design hubs in a MaaS environment
• Turn off fixed route link decision variables
• Limit MOD nodes decisions to hub locations
• New decisions: MOD operators allocated to node,
capacities based on hub capacity
• Turn off profit constraints for public transit
• Consider subsidy only for MOD trips to/from hub(s)
• Output: compare optimal design under common budget/constraint
with existing
17
Transit fixed route extension vs
last mile on-demand feeder
• Pre-req: transit network, zonal demand
• Contribution: transit route network design with on-demand options
• Consider transit route extensions; keep key trunk
lines fixed and let model decide where to switch to
MOD or other smaller capacity bus routes, and type
of MOD
• Explicitly account for transfer impedances
• Bundled fare design
• Output: Redesigned transit network with quantified improvements,
needs for subsidy, predicted ridership, fare bundle pricing
18
Simulation-based algorithm
evaluation in MaaS setting
• Pre-req: existing and proposed fleet mgmt
algorithm
• Contribution: use a fleet simulator (e.g. NOMAD-
RPS) to calibrate the steady state parameters of
each MOD operating cost/travel cost designs
• Output: stable pricing impact of new algorithm,
impact on revenues of other operators, and on
ridership
https://github.com/BUILTNYU/ridepooling
19
Parameter learning/updating
with deployment data
• Pre-req: existing mode/route choice data to estimate external dummy link
utilities per pop segment as logsums, available user route choice data over
time
• Contribution: Bayesian estimation of parameters from new data over time
to refine dummy link utilities, travel cost, inverse optimization of operating
costs from observed stable fare prices
• Output: updated parameters for different population segments over time to
feed back to model use for forecasting new strategic scenarios, digital twin
implementation
20
Alternative (path-based) pricing
schemes
• Pre-req: data from an existing multimodal system with pricing and
ridership for different population segments
• Contribution: zone-based, distance-based, fixed per operator, fixed
per OD (bundle) – optimize for each (for fixed OD, need path-based
pricing) to compare how they impact revenue, consumer surplus,
equity, VMT
• Output: alternative pricing schemes compared to existing system
21
Congestion pricing scenario
analysis
• Pre-req: data from existing multimodal system with steady state
congested travel times, congestion pricing design, estimated
impact of pricing on traffic volumes entering cordon
• Contribution: routes that cross the cordon would charge
additional; dummy links also would increase disutility; MFD
quantifying volume changes on average link travel times inside;
add a destination choice dummy link to not visit
• Output: Sensitivity curve of congestion pricing on operator
revenues, market entry/exit, passing of costs to travelers,
changes to ridership, new adjustments to routes/fleet
sizes/service regions/subsidies
22
Adapting design to freight
deliveries
• Pre-req: data on urban freight/delivery services (order ODs, delivery
modes as MOD services with fleet sizes)
• Contribution: MaaS equilibrium should be applicable to delivery
ecosystems, wait times re-interpreted as schedule delays from time
windows
• Output: evaluation of different MOD providers (e.g. cargo bikes),
hubs/microhubs with transfers from trucks to other modes
23
Conclusion
• MaaS assignment game framework has many rich applications
• Collaborations are welcome!
• Email: [email protected]
24
