Page 1
Introduction Problem Setting Methodology Results Conclusions
Do Self-driving Cars Swallow Public Transport?A Game-theoretical Perspective on Transportation Systems
Nicolas Lanzetti1,2 Gioele Zardini1,2 Maximilian Schiffer1,3 Michael Ostrovsky4 Marco Pavone1
1Autonomous Systems Lab (ASL), Stanford University
2Automatic Control Laboratory (IfA) & Institute for Dynamic Systems and Control (IDSC), ETH Zurich
3TUM School of Management, Technische Universitat Munchen
4Stanford Graduate School of Business, Stanford University
INFORMS Annual Meeting
22nd October, 2019
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 1 of 25
Page 2
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 3
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 4
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 5
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 6
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 7
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 8
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;
• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 9
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;
• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 10
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 11
Introduction Problem Setting Methodology Results Conclusions
Motivation
Challenges
90 hours per yearin congestion
Data from: INRIX, International Parking Institute,Statistical Pocketbook 2018, Aptiv, World Economic Forum, BCG.
30% of congestion caused by driverscircling and struggling for parking
25% CO2
30% Particulate matter60% NOx
Autonomous Mobility-on-Demand Systems
• Ride-hailing fleet of (electric) self-driving cars.
• Controlled by a central operator that
• assigns customer requests to vehicles;• decides on vehicle routes;• rebalances the fleet.
-30% traveltime
-44% parkingplaces
-66%emissions
Data from: Aptiv, World Economic Forum, BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 2 of 25
Page 12
Introduction Problem Setting Methodology Results Conclusions
Motivation
Optimists
• fewer, better utilized vehicles;
• improved pooling, fair matching;
• less congestion, balanced routing.
Pessimists
• increased traffic;
• worsened modal split;
• cannibalization of public transport.
Can autonomous mobility-on-demand (AMoD) systemscannibalize public transport?
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 3 of 25
Page 13
Introduction Problem Setting Methodology Results Conclusions
Motivation
Optimists
• fewer, better utilized vehicles;
• improved pooling, fair matching;
• less congestion, balanced routing.
Pessimists
• increased traffic;
• worsened modal split;
• cannibalization of public transport.
Can AMoD systems cannibalize public transport?
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 3 of 25
Page 14
Introduction Problem Setting Methodology Results Conclusions
Motivation
Optimists
• fewer, better utilized vehicles;
• improved pooling, fair matching;
• less congestion, balanced routing.
Pessimists
• increased traffic;
• worsened modal split;
• cannibalization of public transport.
Can AMoD systems cannibalize public transport?
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 3 of 25
Page 15
Introduction Problem Setting Methodology Results Conclusions
Aims & Scope
Contribution
We present the first algorithmic framework that
• captures the dynamics between multiple mobility service providers and customers;
• considers constraints of a complex real-world transportation network; and
• allows for multimodal customer decisions.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 4 of 25
Page 16
Introduction Problem Setting Methodology Results Conclusions
Aims & Scope
Contribution
We present the first algorithmic framework that
• captures the dynamics between multiple mobility service providers and customers;
• considers constraints of a complex real-world transportation network; and
• allows for multimodal customer decisions.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 4 of 25
Page 17
Introduction Problem Setting Methodology Results Conclusions
Aims & Scope
Contribution
We present the first algorithmic framework that
• captures the dynamics between multiple mobility service providers and customers;
• considers constraints of a complex real-world transportation network; and
• allows for multimodal customer decisions.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 4 of 25
Page 18
Introduction Problem Setting Methodology Results Conclusions
Problem Setting – Who is playing?
Stakeholder Role Goal
Mobility Service Providers Offer mobility services Profit
Municipalities Offer mobility services Social welfare
Customers Request mobility services Individual benefit
vs.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 5 of 25
Page 19
Introduction Problem Setting Methodology Results Conclusions
Problem Setting – Who is playing?
Stakeholder Role Goal
Mobility Service Providers Offer mobility services Profit
Municipalities Offer mobility services Social welfare
Customers Request mobility services Individual benefit
vs.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 5 of 25
Page 20
Introduction Problem Setting Methodology Results Conclusions
Problem Setting – Who is playing?
Stakeholder Role Goal
Mobility Service Providers Offer mobility services Profit
Municipalities Offer mobility services Social welfare
Customers Request mobility services Individual benefit
vs.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 5 of 25
Page 21
Introduction Problem Setting Methodology Results Conclusions
Problem Setting – Who is playing?
Stakeholder Role Goal
Mobility Service Providers Offer mobility services Profit
Municipalities Offer mobility services Social welfare
Customers Request mobility services Individual benefit
vs.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 5 of 25
Page 22
Introduction Problem Setting Methodology Results Conclusions
Problem Setting – Who is playing?
Stakeholder Role Goal
Mobility Service Providers Offer mobility services Profit
Municipalities Offer mobility services Social welfare
Customers Request mobility services Individual benefit
vs.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 5 of 25
Page 23
Introduction Problem Setting Methodology Results Conclusions
Problem Setting – A Two-level System
Customers
PTAMSPGame Theory
Transportation Market Place
Transportation ResearchTransportation Network
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 6 of 25
Page 24
Introduction Problem Setting Methodology Results Conclusions
Problem Setting – A Two-level System
Customers
PTAMSPGame Theory
Transportation Market Place
Transportation ResearchTransportation Network
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 6 of 25
Page 25
Introduction Problem Setting Methodology Results Conclusions
Modeling – Graph Representation
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 7 of 25
Page 26
Introduction Problem Setting Methodology Results Conclusions
Modeling – Graph Representation
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 7 of 25
Page 27
Introduction Problem Setting Methodology Results Conclusions
Modeling – Graph Representation
G0: Free Subgraph
G1: Subgraph controlled by operator 1
G2: Subgraph controlled by operator 2
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 7 of 25
Page 28
Introduction Problem Setting Methodology Results Conclusions
Modeling – Graph Representation
G0: Free Subgraph
G1: Subgraph controlled by operator 1
G2: Subgraph controlled by operator 2
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 7 of 25
Page 29
Introduction Problem Setting Methodology Results Conclusions
Modeling – Graph Representation
G0: Free Subgraph
G1: Subgraph controlled by operator 1
G2: Subgraph controlled by operator 2
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 7 of 25
Page 30
Introduction Problem Setting Methodology Results Conclusions
Modeling – Graph Representation
G0: Free Subgraph
G1: Subgraph controlled by operator 1
G2: Subgraph controlled by operator 2
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 7 of 25
Page 31
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers may move:
1. on the “free subgraph” G0, and
2. on the fully-connected operators’ subgraphs G1, . . . ,GNo .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 8 of 25
Page 32
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers may move:
1. on the “free subgraph” G0, and
2. on the fully-connected operators’ subgraphs G1, . . . ,GNo .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 8 of 25
Page 33
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers may move:
1. on the “free subgraph” G0, and
2. on the fully-connected operators’ subgraphs G1, . . . ,GNo .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 8 of 25
Page 34
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers may move:
1. on the “free subgraph” G0, and
2. on the fully-connected operators’ subgraphs G1, . . . ,GNo .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 8 of 25
Page 35
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers’ Route Decision
Select a reaction curve φi :
φi (p) = α ≡α customers per unit
time on path p
with related cost Ji (φi , pr1, . . . , prNo).
Remark (Requirements for φi )
1. Demand conservation: φi ∈ Φ(di ).
2. Feasibility: φi ∈ Ac,i .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 9 of 25
Page 36
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers’ Route Decision
Select a reaction curve φi :
φi (p) = α ≡α customers per unit
time on path p
with related cost Ji (φi , pr1, . . . , prNo).
Remark (Requirements for φi )
1. Demand conservation: φi ∈ Φ(di ).
2. Feasibility: φi ∈ Ac,i .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 9 of 25
Page 37
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers’ Route Decision
Select a reaction curve φi :
φi (p) = α ≡α customers per unit
time on path p
with related cost Ji (φi , pr1, . . . , prNo).
Remark (Requirements for φi )
1. Demand conservation: φi ∈ Φ(di ).
2. Feasibility: φi ∈ Ac,i .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 9 of 25
Page 38
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers’ Route Decision
Select a reaction curve φi :
φi (p) = α ≡α customers per unit
time on path p
with related cost Ji (φi , pr1, . . . , prNo).
Remark (Requirements for φi )
1. Demand conservation: φi ∈ Φ(di ).
2. Feasibility: φi ∈ Ac,i .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 9 of 25
Page 39
Introduction Problem Setting Methodology Results Conclusions
Modeling – Customers
Customers’ Route Decision
Select a reaction curve φi :
φi (p) = α ≡α customers per unit
time on path p
with related cost Ji (φi , pr1, . . . , prNo).
Remark (Requirements for φi )
1. Demand conservation: φi ∈ Φ(di ).
2. Feasibility: φi ∈ Ac,i .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 9 of 25
Page 40
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
1. Select a pricing strategy pr ∈ Pr:
pr : Vj × Vj → R≥0 ∪ {+∞}(o, d) 7→ price.
2. Serve each demand i with some flowsFi = {f1
i , . . . , fLi
i }.3. Rebalance the system with some flows
F0 = {f10, . . . , f
L00 }.
Remark (Requirements for the flows)
1. Demand satisfaction: Fi ∈ Hi (φi ).
2. Feasibility: (F1, . . . ,FM ,F0) ∈ Ao,i .
2
4
1
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 10 of 25
Page 41
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
1. Select a pricing strategy pr ∈ Pr:
pr : Vj × Vj → R≥0 ∪ {+∞}(o, d) 7→ price.
2. Serve each demand i with some flowsFi = {f1
i , . . . , fLi
i }.3. Rebalance the system with some flows
F0 = {f10, . . . , f
L00 }.
Remark (Requirements for the flows)
1. Demand satisfaction: Fi ∈ Hi (φi ).
2. Feasibility: (F1, . . . ,FM ,F0) ∈ Ao,i .
2
4
1
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 10 of 25
Page 42
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
1. Select a pricing strategy pr ∈ Pr:
pr : Vj × Vj → R≥0 ∪ {+∞}(o, d) 7→ price.
2. Serve each demand i with some flowsFi = {f1
i , . . . , fLi
i }.3. Rebalance the system with some flows
F0 = {f10, . . . , f
L00 }.
Remark (Requirements for the flows)
1. Demand satisfaction: Fi ∈ Hi (φi ).
2. Feasibility: (F1, . . . ,FM ,F0) ∈ Ao,i .
2
4
1
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 10 of 25
Page 43
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
1. Select a pricing strategy pr ∈ Pr:
pr : Vj × Vj → R≥0 ∪ {+∞}(o, d) 7→ price.
2. Serve each demand i with some flowsFi = {f1
i , . . . , fLi
i }.
3. Rebalance the system with some flowsF0 = {f1
0, . . . , fL00 }.
Remark (Requirements for the flows)
1. Demand satisfaction: Fi ∈ Hi (φi ).
2. Feasibility: (F1, . . . ,FM ,F0) ∈ Ao,i .
2
4
1
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 10 of 25
Page 44
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
1. Select a pricing strategy pr ∈ Pr:
pr : Vj × Vj → R≥0 ∪ {+∞}(o, d) 7→ price.
2. Serve each demand i with some flowsFi = {f1
i , . . . , fLi
i }.3. Rebalance the system with some flows
F0 = {f10, . . . , f
L00 }.
Remark (Requirements for the flows)
1. Demand satisfaction: Fi ∈ Hi (φi ).
2. Feasibility: (F1, . . . ,FM ,F0) ∈ Ao,i .
2
4
1
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 10 of 25
Page 45
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
1. Select a pricing strategy pr ∈ Pr:
pr : Vj × Vj → R≥0 ∪ {+∞}(o, d) 7→ price.
2. Serve each demand i with some flowsFi = {f1
i , . . . , fLi
i }.3. Rebalance the system with some flows
F0 = {f10, . . . , f
L00 }.
Remark (Requirements for the flows)
1. Demand satisfaction: Fi ∈ Hi (φi ).
2. Feasibility: (F1, . . . ,FM ,F0) ∈ Ao,i .
2
4
1
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 10 of 25
Page 46
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
Operators’ Profit Maximization
Revenuej :=M∑i=1
∑p∈S(di )
∑a∈p,
a∈Aj
φi (p) · prj(sj(a), tj(a))
Costj :=
minFi∈Hi (φi ),
F0∈2F(Gj ),
({Fi}Mi=1,F0)∈Ao,j
M∑i=1
cj(Fi ) + cj(F0)
HenceUj(prj , {φi}Mi=1) := Revenuej − Costj
Rate Price
Cost serving
demand i
Cost rebalancing
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 11 of 25
Page 47
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
Operators’ Profit Maximization
Revenuej :=M∑i=1
∑p∈S(di )
∑a∈p,
a∈Aj
φi (p) · prj(sj(a), tj(a))
Costj :=
minFi∈Hi (φi ),
F0∈2F(Gj ),
({Fi}Mi=1,F0)∈Ao,j
M∑i=1
cj(Fi ) + cj(F0)
HenceUj(prj , {φi}Mi=1) := Revenuej − Costj
Rate Price
Cost serving
demand i
Cost rebalancing
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 11 of 25
Page 48
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
Operators’ Profit Maximization
Revenuej :=M∑i=1
∑p∈S(di )
∑a∈p,
a∈Aj
φi (p) · prj(sj(a), tj(a))
Costj := minFi∈Hi (φi ),
F0∈2F(Gj ),
({Fi}Mi=1,F0)∈Ao,j
M∑i=1
cj(Fi ) + cj(F0)
HenceUj(prj , {φi}Mi=1) := Revenuej − Costj
Rate Price
Cost serving
demand i
Cost rebalancing
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 11 of 25
Page 49
Introduction Problem Setting Methodology Results Conclusions
Modeling – Operators
Operators’ Profit Maximization
Revenuej :=M∑i=1
∑p∈S(di )
∑a∈p,
a∈Aj
φi (p) · prj(sj(a), tj(a))
Costj := minFi∈Hi (φi ),
F0∈2F(Gj ),
({Fi}Mi=1,F0)∈Ao,j
M∑i=1
cj(Fi ) + cj(F0)
HenceUj(prj , {φi}Mi=1) := Revenuej − Costj
Rate Price
Cost serving
demand i
Cost rebalancing
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 11 of 25
Page 50
Introduction Problem Setting Methodology Results Conclusions
Customers Equilibrium
Definition (Customer Equilibrium)
The reaction curve φ?i is an equilibrium for the demand di if
Ji (φ?i , pr1, . . . , prNo
) ≤ Ji (φi , pr1, . . . , prNo) ∀φi ∈ Φ(di ) ∩ Ac,i
The set of equilibria is Ei (pr1, . . . , prNo).
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 12 of 25
Page 51
Introduction Problem Setting Methodology Results Conclusions
Game Equilibrium
Definition (Game equilibrium)
The reaction curves and the pricing strategies ({φ?i }Mi=1, {pr?j }No
j=1) ∈∏M
i=1 Φ(di ) ∩ Ac,i ×∏No
j=1 Prj are an equilibrium if
1. the customers are at equilibrium, and
2. no operator can increase her profit by unilaterally deviating from her pricing strategy.
Formally, ({φ?i }Mi=1, {pr?j }No
j=1) is a equilibrium if
1. for all i ∈ {1, . . . ,M}φ?i ∈ Ei (pr?1 , . . . , pr?No
).
2. for all j ∈ {1, . . . ,No}
Uj(pr?j , {Ei (pr?1 , . . . , pr?No)}Mi=1) ≥ Uj(prj , {Ei (pr?1 , . . . , prj , . . . , pr?No
)}Mi=1), ∀ prj ∈ Prj .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 13 of 25
Page 52
Introduction Problem Setting Methodology Results Conclusions
Game Equilibrium
Definition (Game equilibrium)
The reaction curves and the pricing strategies ({φ?i }Mi=1, {pr?j }No
j=1) ∈∏M
i=1 Φ(di ) ∩ Ac,i ×∏No
j=1 Prj are an equilibrium if
1. the customers are at equilibrium, and
2. no operator can increase her profit by unilaterally deviating from her pricing strategy.
Formally, ({φ?i }Mi=1, {pr?j }No
j=1) is a equilibrium if
1. for all i ∈ {1, . . . ,M}φ?i ∈ Ei (pr?1 , . . . , pr?No
).
2. for all j ∈ {1, . . . ,No}
Uj(pr?j , {Ei (pr?1 , . . . , pr?No)}Mi=1) ≥ Uj(prj , {Ei (pr?1 , . . . , prj , . . . , pr?No
)}Mi=1), ∀ prj ∈ Prj .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 13 of 25
Page 53
Introduction Problem Setting Methodology Results Conclusions
Game Equilibrium
Definition (Game equilibrium)
The reaction curves and the pricing strategies ({φ?i }Mi=1, {pr?j }No
j=1) ∈∏M
i=1 Φ(di ) ∩ Ac,i ×∏No
j=1 Prj are an equilibrium if
1. the customers are at equilibrium, and
2. no operator can increase her profit by unilaterally deviating from her pricing strategy.
Formally, ({φ?i }Mi=1, {pr?j }No
j=1) is a equilibrium if
1. for all i ∈ {1, . . . ,M}φ?i ∈ Ei (pr?1 , . . . , pr?No
).
2. for all j ∈ {1, . . . ,No}
Uj(pr?j , {Ei (pr?1 , . . . , pr?No)}Mi=1) ≥ Uj(prj , {Ei (pr?1 , . . . , prj , . . . , pr?No
)}Mi=1), ∀ prj ∈ Prj .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 13 of 25
Page 54
Introduction Problem Setting Methodology Results Conclusions
Game Equilibrium
Definition (Game equilibrium)
The reaction curves and the pricing strategies ({φ?i }Mi=1, {pr?j }No
j=1) ∈∏M
i=1 Φ(di ) ∩ Ac,i ×∏No
j=1 Prj are an equilibrium if
1. the customers are at equilibrium, and
2. no operator can increase her profit by unilaterally deviating from her pricing strategy.
Formally, ({φ?i }Mi=1, {pr?j }No
j=1) is a equilibrium if
1. for all i ∈ {1, . . . ,M}φ?i ∈ Ei (pr?1 , . . . , pr?No
).
2. for all j ∈ {1, . . . ,No}
Uj(pr?j , {Ei (pr?1 , . . . , pr?No)}Mi=1) ≥ Uj(prj , {Ei (pr?1 , . . . , prj , . . . , pr?No
)}Mi=1), ∀ prj ∈ Prj .
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 13 of 25
Page 55
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Settings
Players:
• M demands
• Two operators:
Name Graph Pricing Strategies Set
Operator 1 AMoD System G1 Pr1 = RV1×V1
≥0 ≡ All nonnegative functions
Operator 2 PTA/Municipality G2 Pr2 = {pr2}
Assumptions
• Time-invariant setting.
• The time from o to d through path p is known a priori.
• Multimodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 14 of 25
Page 56
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Settings
Players:
• M demands
• Two operators:
Name Graph Pricing Strategies Set
Operator 1 AMoD System G1 Pr1 = RV1×V1
≥0 ≡ All nonnegative functions
Operator 2 PTA/Municipality G2 Pr2 = {pr2}
Assumptions
• Time-invariant setting.
• The time from o to d through path p is known a priori.
• Multimodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 14 of 25
Page 57
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Settings
Players:
• M demands
• Two operators:
Name Graph Pricing Strategies Set
Operator 1 AMoD System G1 Pr1 = RV1×V1
≥0 ≡ All nonnegative functions
Operator 2 PTA/Municipality G2 Pr2 = {pr2}
Assumptions
• Time-invariant setting.
• The time from o to d through path p is known a priori.
• Multimodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 14 of 25
Page 58
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Settings
Players:
• M demands
• Two operators:
Name Graph Pricing Strategies Set
Operator 1 AMoD System G1 Pr1 = RV1×V1
≥0 ≡ All nonnegative functions
Operator 2 PTA/Municipality G2 Pr2 = {pr2}
Assumptions
• Time-invariant setting.
• The time from o to d through path p is known a priori.
• Multimodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 14 of 25
Page 59
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Settings
Players:
• M demands
• Two operators:
Name Graph Pricing Strategies Set
Operator 1 AMoD System G1 Pr1 = RV1×V1
≥0 ≡ All nonnegative functions
Operator 2 PTA/Municipality G2 Pr2 = {pr2}
Assumptions
• Time-invariant setting.
• The time from o to d through path p is known a priori.
• Multimodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 14 of 25
Page 60
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Settings
Players:
• M demands
• Two operators:
Name Graph Pricing Strategies Set
Operator 1 AMoD System G1 Pr1 = RV1×V1
≥0 ≡ All nonnegative functions
Operator 2 PTA/Municipality G2 Pr2 = {pr2}
Assumptions
• Time-invariant setting.
• The time from o to d through path p is known a priori.
• Multimodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 14 of 25
Page 61
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Customers
• Multimodal choice:
pAMoD,i : AMoD path⇒ Ac,i = {φ |φ(p) = 0 ∀ p 6= pAMoD,i , pPT,i}
pPT,i : public transport and walking path
• Monetary costs of fares and time:
Ji (φ, pr1, pr2) = (pr1(o, d) + VT · tAMoD,i ) · φ(pAMoD,i ) + (prPT,i + VT · tPT,i ) · φ(pPT,i ).
Equilibrium
φi =
EVT
[
arg minφ∈Φ(di )∩Ac,i
Ji (φ, pr1, pr2)
]
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 15 of 25
Page 62
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Customers
• Multimodal choice:
pAMoD,i : AMoD path
⇒ Ac,i = {φ |φ(p) = 0 ∀ p 6= pAMoD,i , pPT,i}pPT,i : public transport and walking path
• Monetary costs of fares and time:
Ji (φ, pr1, pr2) = (pr1(o, d) + VT · tAMoD,i ) · φ(pAMoD,i ) + (prPT,i + VT · tPT,i ) · φ(pPT,i ).
Equilibrium
φi =
EVT
[
arg minφ∈Φ(di )∩Ac,i
Ji (φ, pr1, pr2)
]
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 15 of 25
Page 63
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Customers
• Multimodal choice:
pAMoD,i : AMoD path
⇒ Ac,i = {φ |φ(p) = 0 ∀ p 6= pAMoD,i , pPT,i}
pPT,i : public transport and walking path
• Monetary costs of fares and time:
Ji (φ, pr1, pr2) = (pr1(o, d) + VT · tAMoD,i ) · φ(pAMoD,i ) + (prPT,i + VT · tPT,i ) · φ(pPT,i ).
Equilibrium
φi =
EVT
[
arg minφ∈Φ(di )∩Ac,i
Ji (φ, pr1, pr2)
]
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 15 of 25
Page 64
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Customers
• Multimodal choice:
pAMoD,i : AMoD path⇒ Ac,i = {φ |φ(p) = 0 ∀ p 6= pAMoD,i , pPT,i}
pPT,i : public transport and walking path
• Monetary costs of fares and time:
Ji (φ, pr1, pr2) = (pr1(o, d) + VT · tAMoD,i ) · φ(pAMoD,i ) + (prPT,i + VT · tPT,i ) · φ(pPT,i ).
Equilibrium
φi =
EVT
[
arg minφ∈Φ(di )∩Ac,i
Ji (φ, pr1, pr2)
]
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 15 of 25
Page 65
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Customers
• Multimodal choice:
pAMoD,i : AMoD path⇒ Ac,i = {φ |φ(p) = 0 ∀ p 6= pAMoD,i , pPT,i}
pPT,i : public transport and walking path
• Monetary costs of fares and time:
Ji (φ, pr1, pr2) = (pr1(o, d) + VT · tAMoD,i ) · φ(pAMoD,i ) + (prPT,i + VT · tPT,i ) · φ(pPT,i ).
Equilibrium
φi =
EVT
[
arg minφ∈Φ(di )∩Ac,i
Ji (φ, pr1, pr2)
]
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 15 of 25
Page 66
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Customers
• Multimodal choice:
pAMoD,i : AMoD path⇒ Ac,i = {φ |φ(p) = 0 ∀ p 6= pAMoD,i , pPT,i}
pPT,i : public transport and walking path
• Monetary costs of fares and time:
Ji (φ, pr1, pr2) = (pr1(o, d) + VT · tAMoD,i ) · φ(pAMoD,i ) + (prPT,i + VT · tPT,i ) · φ(pPT,i ).
Equilibrium
φi =
EVT
[
arg minφ∈Φ(di )∩Ac,i
Ji (φ, pr1, pr2)
]
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 15 of 25
Page 67
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – Customers
• Multimodal choice:
pAMoD,i : AMoD path⇒ Ac,i = {φ |φ(p) = 0 ∀ p 6= pAMoD,i , pPT,i}
pPT,i : public transport and walking path
• Monetary costs of fares and time:
Ji (φ, pr1, pr2) = (pr1(o, d) + VT · tAMoD,i ) · φ(pAMoD,i ) + (prPT,i + VT · tPT,i ) · φ(pPT,i ).
Equilibrium
φi = EVT
[arg min
φ∈Φ(di )∩Ac,i
Ji (φ, pr1, pr2)
]
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 15 of 25
Page 68
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – AMoD Operator
• She assigns requests to vehicles (selecting flows).
• Vehicles must be conserved and are limited:
Ao,1 ={
(F1, . . . ,FM ,F0)∣∣∣ (F1, . . . ,FM ,F0) is balanced ∧ number of cars ≤ Nveh
}.
• Vehicles flow F = {f1, . . . , fN} cost:
co,1(F) =∑f∈F
χrate(f )∑
a∈χpath(f )
cd,1(a)
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 16 of 25
Page 69
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – AMoD Operator
• She assigns requests to vehicles (selecting flows).
• Vehicles must be conserved and are limited:
Ao,1 ={
(F1, . . . ,FM ,F0)∣∣∣ (F1, . . . ,FM ,F0) is balanced ∧ number of cars ≤ Nveh
}.
• Vehicles flow F = {f1, . . . , fN} cost:
co,1(F) =∑f∈F
χrate(f )∑
a∈χpath(f )
cd,1(a)
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 16 of 25
Page 70
Introduction Problem Setting Methodology Results Conclusions
AMoD Framework – AMoD Operator
• She assigns requests to vehicles (selecting flows).
• Vehicles must be conserved and are limited:
Ao,1 ={
(F1, . . . ,FM ,F0)∣∣∣ (F1, . . . ,FM ,F0) is balanced ∧ number of cars ≤ Nveh
}.
• Vehicles flow F = {f1, . . . , fN} cost:
co,1(F) =∑f∈F
χrate(f )∑
a∈χpath(f )
cd,1(a)
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 16 of 25
Page 71
Introduction Problem Setting Methodology Results Conclusions
Equilibrium
Theorem (Equilibrium)
If customers have a uniformly distributed value of time, then:
• The game has a (possibly non-unique) equilibrium.
• Consider the reaction curves {φ?i }Mi=1 and the pricing strategies pr?1 and pr?2 such that
1. pr?1 (o, d) = 0 if there is no demand from o to d ;2. pr?1 (o, d) = p? where p? is the solution of a convex quadratic program;3. pr?2 (o, d) = pr2(o, d);4. φ?
i ∈ Ei (pr?1 , pr?2 ).
Then, ({φ?i }Mi=1, pr?1 , pr?2) is an equilibrium.
• All equilibria result in the same profit and customers’ reaction curves.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 17 of 25
Page 72
Introduction Problem Setting Methodology Results Conclusions
Equilibrium
Theorem (Equilibrium)
If customers have a uniformly distributed value of time, then:
• The game has a (possibly non-unique) equilibrium.
• Consider the reaction curves {φ?i }Mi=1 and the pricing strategies pr?1 and pr?2 such that
1. pr?1 (o, d) = 0 if there is no demand from o to d ;2. pr?1 (o, d) = p? where p? is the solution of a convex quadratic program;3. pr?2 (o, d) = pr2(o, d);4. φ?
i ∈ Ei (pr?1 , pr?2 ).
Then, ({φ?i }Mi=1, pr?1 , pr?2) is an equilibrium.
• All equilibria result in the same profit and customers’ reaction curves.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 17 of 25
Page 73
Introduction Problem Setting Methodology Results Conclusions
Equilibrium
Theorem (Equilibrium)
If customers have a uniformly distributed value of time, then:
• The game has a (possibly non-unique) equilibrium.
• Consider the reaction curves {φ?i }Mi=1 and the pricing strategies pr?1 and pr?2 such that
1. pr?1 (o, d) = 0 if there is no demand from o to d ;2. pr?1 (o, d) = p? where p? is the solution of a convex quadratic program;3. pr?2 (o, d) = pr2(o, d);4. φ?
i ∈ Ei (pr?1 , pr?2 ).
Then, ({φ?i }Mi=1, pr?1 , pr?2) is an equilibrium.
• All equilibria result in the same profit and customers’ reaction curves.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 17 of 25
Page 74
Introduction Problem Setting Methodology Results Conclusions
Equilibrium
Theorem (Equilibrium)
If customers have a uniformly distributed value of time, then:
• The game has a (possibly non-unique) equilibrium.
• Consider the reaction curves {φ?i }Mi=1 and the pricing strategies pr?1 and pr?2 such that
1. pr?1 (o, d) = 0 if there is no demand from o to d ;2. pr?1 (o, d) = p? where p? is the solution of a convex quadratic program;3. pr?2 (o, d) = pr2(o, d);4. φ?
i ∈ Ei (pr?1 , pr?2 ).
Then, ({φ?i }Mi=1, pr?1 , pr?2) is an equilibrium.
• All equilibria result in the same profit and customers’ reaction curves.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 17 of 25
Page 75
Introduction Problem Setting Methodology Results Conclusions
Case Study – Berlin, Germany (∼ 9,000 requests, evening peak)
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 18 of 25
Page 76
Introduction Problem Setting Methodology Results Conclusions
Results – Base Case (fleet of ∼ 8,000 vehicles)
42.3%
49.3%
8.4%AMoD
Public transport
Walking
0
10
20
30
40
50
60
70
80
90
100
Pro
fit
of
atr
ip/
Rev
enu
eo
fa
trip
[%]
Approx. equal modal splitamong AMoD and public transport.
Most AMoD trips yield a high profit.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 19 of 25
Page 77
Introduction Problem Setting Methodology Results Conclusions
Results – Base Case (fleet of ∼ 8,000 vehicles)
42.3%
49.3%
8.4%AMoD
Public transport
Walking
0
10
20
30
40
50
60
70
80
90
100
Pro
fit
of
atr
ip/
Rev
enu
eo
fa
trip
[%]
Approx. equal modal splitamong AMoD and public transport. Most AMoD trips yield a high profit.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 19 of 25
Page 78
Introduction Problem Setting Methodology Results Conclusions
Results – Base Case (fleet of ∼ 8,000 vehicles)
468
101214161820
Tri
pco
st[U
SD
]
525 526 527 528 529 530 531 5320
20
40
60
80
100
Demand ID/Trip [–]
Sh
are
[%]
At microscopic level,the modal split appears
to be less balanced.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 20 of 25
Page 79
Introduction Problem Setting Methodology Results Conclusions
Results – Sensitivity of the Equilibrium
AMoD Operator
1. Different vehicles
2. Larger fleet size
3. Heterogenous prices
Municipality
1. Lower public transport prices
2. More efficient public transportinfrastructure
3. AMoD service tax
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 21 of 25
Page 80
Introduction Problem Setting Methodology Results Conclusions
Results – Sensitivity of the Equilibrium
AMoD Operator
1. Different vehicles
2. Larger fleet size
3. Heterogenous prices
Municipality
1. Lower public transport prices
2. More efficient public transportinfrastructure
3. AMoD service tax
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 21 of 25
Page 81
Introduction Problem Setting Methodology Results Conclusions
Results – Sensitivity of the Equilibrium
AMoD Operator
1. Different vehicles
2. Larger fleet size
3. Heterogenous prices
Municipality
1. Lower public transport prices
2. More efficient public transportinfrastructure
3. AMoD service tax
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 21 of 25
Page 82
Introduction Problem Setting Methodology Results Conclusions
Results – Vehicles
Modal shareAMoD
ProfitAMoD
RevenueAMoD
RevenueMunicipality
0.001
0.01
0.1
1
a
Nor
ma
lize
dva
lue Nominal 0.34 USD/km
Non-autonomous (Low-wage) 1.83 USD/km
Non-autonomous (High-wage) 3.26 USD/km
Non-electrified 0.36 USD/km
Data from:
Cost-based analysis of autonomous mobility services [Bosch et al., 2017]
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 22 of 25
Page 83
Introduction Problem Setting Methodology Results Conclusions
Results – Public Transport Price
0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 5 5,5 60
10
20
30
40
50
60
70
80
90
100
Public transport price [USD]
Mo
da
lsh
are
[%]
0
20
40
Pro
fit,
Rev
enu
e[U
SD/
s]
AMoD
Public transport
Walking
Profit AMoD
Revenue AMoD
Revenue municipality
Today
A free public transportcounteracts
the AMoD system.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 23 of 25
Page 84
Introduction Problem Setting Methodology Results Conclusions
Results – AMoD Service Tax
0 10 20 30 40 50 60 70 800
10
20
30
40
50
60
70
80
90
100
Service tax [%]
Mo
da
lsh
are
[%]
0
20
40
Pro
fit,
Rev
enu
e[U
SD/
s]
AMoD
Public transport
Walking
Profit AMoD
Revenue AMoD
Revenue municipality
Only significanttaxes decrease
the modal share.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 24 of 25
Page 85
Introduction Problem Setting Methodology Results Conclusions
Conclusions
Summary
• General game-theoretical framework for transportation systems.
• Specific framework for an AMoD system competing with the public transport.
• In our case study, the AMoD system attracts 42% of the customers.
Managerial Insights
• Vehicles autonomy significantly affects the equilibrium.
• A free public transportation service counteracts the AMoD operator.
• Imposing high taxes on an AMoD system can impact the modal split.
Outlook
• Competition between multiple AMoD operators.
• Intermodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 25 of 25
Page 86
Introduction Problem Setting Methodology Results Conclusions
Conclusions
Summary
• General game-theoretical framework for transportation systems.
• Specific framework for an AMoD system competing with the public transport.
• In our case study, the AMoD system attracts 42% of the customers.
Managerial Insights
• Vehicles autonomy significantly affects the equilibrium.
• A free public transportation service counteracts the AMoD operator.
• Imposing high taxes on an AMoD system can impact the modal split.
Outlook
• Competition between multiple AMoD operators.
• Intermodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 25 of 25
Page 87
Introduction Problem Setting Methodology Results Conclusions
Conclusions
Summary
• General game-theoretical framework for transportation systems.
• Specific framework for an AMoD system competing with the public transport.
• In our case study, the AMoD system attracts 42% of the customers.
Managerial Insights
• Vehicles autonomy significantly affects the equilibrium.
• A free public transportation service counteracts the AMoD operator.
• Imposing high taxes on an AMoD system can impact the modal split.
Outlook
• Competition between multiple AMoD operators.
• Intermodal route selection.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 25 of 25
Page 88
Backup Slides
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 26 of 25
Page 89
Backup Slides
References
• Time in traffic: INRIX.
• Congestion: International Parking Institute (IPI) 2012 Emerging Trends in Parking Study.
• Emissions: Statistical pocketbook 2018.
• Benefits autonomous vehicles: Aptiv, World Economic Forum, and BCG.
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 27 of 25
Page 90
Backup Slides
Case Study – Data
Road Network: OpenStreetMap.
Public Transit Network: GTFS (topology and travel time).
Origin-destination pairs: MatSim scenario Berlin (scaled with a factor 10).
Considered area: We have:
• 16 km× 16 km,• 9052 travel requests (12.8 travel demands per second).
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 28 of 25
Page 91
Backup Slides
Case Study – Parameters
Parameter Value
Public transit price 3.12 USD
Value of time minimum 10 USD/h
Value of time maximum 17 USD/h
Operation cost 0.34 USD/km
Walking velocity 1.4 m/s
Average wait S-Bahn/U-Bahn 2.5 min
Average wait tram 3.5 min
Average wait bus 5 min
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 29 of 25
Page 92
Backup Slides
Case Study – Fleet Size
City Number of registered cars Number of taxi licenses Percentage
Berlin 1,344,000 8,373 0.6%
New York City 3,000,000 13,237 0.4%
San Francisco 494,000 1,800 0.4%
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 30 of 25
Page 93
Backup Slides
Results – Fleet Size
1 3 5 7 9 11 13 15 17 190
10
20
30
40
50
60
70
80
90
100
Fleet size [1000 vehicles]
Mo
da
lsh
are
[%]
0
20
40
60
Pro
fit,
Rev
enu
e[U
SD/
s]
AMoD
Public transport
Walking
Profit AMoD
Revenue AMoD
Revenue municipality
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 31 of 25
Page 94
Backup Slides
Results – Customers Heterogenity
Change
Profit AMoD +0.3%
AMoD modal share −0.4%
Revenue Municipality +0.1%
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 32 of 25
Page 95
Backup Slides
Results – Public Transit Infrastructure
1 1,25 1,5 1,75 2 2,25 2,5 2,75 30
10
20
30
40
50
60
70
80
90
100
Public transport frequency scaling [–]
Mo
da
lsh
are
[%]
0
10
20
30
40
Pro
fit,
Rev
enu
e[U
SD/
s]
AMoD
Public transport
Walking
Profit AMoD
Revenue AMoD
Revenue municipality
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 33 of 25
Page 96
Backup Slides
Results – AMoD Service Tax
0 10 20 30 40 50 60 70 800
20
40
Service tax [%]
Rev
enu
em
un
icip
ality
[US
D/
s]
Fares paid the customers
Tax paid by the AMoD operator
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 34 of 25
Page 97
Backup Slides
Results – AMoD Service Tax
0 10 20 30 40 50 60 70 800
10
20
30
40
50
60
70
80
90
100
Service tax [%]
Mo
da
lsh
are
[%]
0
20
40
Pro
fit,
Rev
enu
e[U
SD/
s]
AMoD
Public transport
Walking
Profit AMoD
Revenue AMoD
Revenue municipality
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 35 of 25
Page 98
Backup Slides
Do Self-driving Cars Swallow Public Transport?22nd October, 2019 | Nicolas Lanzetti | [email protected] 36 of 25