+ All Categories
Home > Government & Nonprofit > SMART Seminar Series: Mechanism Design for Ridesharing

SMART Seminar Series: Mechanism Design for Ridesharing

Date post: 01-Jul-2015
Category:
Upload: smart-infrastructure-facility
View: 224 times
Download: 0 times
Share this document with a friend
Description:
SMART Infrastructure Facility was pleased to host Dr Dengji Zhao, a Research Fellow at the University of Southampton, UK as he presented ‘Mechanism Design for Ridesharing’ as part of the SMART Seminar Series on October 16th, 2014.
42
The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion Mechanism Design for Ridesharing Dengji Zhao 1 Dongmo Zhang 2 Enrico Gerding 1 Gopal Ramchurn 1 Makoto Yokoo 3 David Parkes 4 Nick Jennings 1 1 University of Southampton, UK 2 University of Western Sydney, Australia 3 Kyushu University, Japan 4 Harvard University, US 1 / 28
Transcript
Page 1: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Mechanism Design for Ridesharing

Dengji Zhao1 Dongmo Zhang2 Enrico Gerding1

Gopal Ramchurn1 Makoto Yokoo3 David Parkes4

Nick Jennings1

1University of Southampton, UK2University of Western Sydney, Australia

3Kyushu University, Japan4Harvard University, US

1 / 28

Page 2: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Ridesharing Example

2 / 28

Page 3: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Ridesharing Example

2 / 28

Page 4: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Ridesharing Example

2 / 28

Page 5: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Ridesharing Example

Questions:How to arrange thesharing?How much shouldthey pay/receive?

2 / 28

Page 6: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

History

Began in the 1940s in NorthAmericaBeen promoted because of

fuel shortages, air pollution andtraffic congestion

Peaked in the US in 1970 with acommute mode share of 20.4%

3 / 28

Page 7: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Public and Private Promotions

4 / 28

Page 8: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

People are still NOT well motivated!

not going well...

Australia (Queensland) will end ridesharing lanesThe average car carries just 1.6 people

5 / 28

Page 9: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

What are the obstacles?

Safety and PrivacyFlexibility and Reliability...

6 / 28

Page 10: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

What are the obstacles?

Safety and PrivacyFlexibility and Reliability...Complicated join proceduresNo free market competition!

6 / 28

Page 11: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

What we can do?

Use Mechanism Design to build ridesharing:Automated ride matchingAutomated (profitable) price setting

7 / 28

Page 12: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

What we can do?

Use Mechanism Design to build ridesharing:Automated ride matchingAutomated (profitable) price setting

(A -> B, cost: 10)

Driver 1

Exisitng

Ridesharing

Services

Auction

Based

Ridesharing

post trips

(A -> B, cost: 15)

Rider 2search trips

3

ride arrangement

(A -> B, cost: 10)

Driver

(A -> B, cost: 15)

Rider

1

1

post trips

post trips

2

ride schedules

2

7 / 28

Page 13: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

What we can do?

Use Mechanism Design to build ridesharing:Automated ride matchingAutomated (profitable) price setting

to answer...Questions:

How to arrange the sharing?How much should commuters pay/receive?

7 / 28

Page 14: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Outline

1 The Model

2 Fixed-price Mechanisms

3 VCG with Reserve Prices

4 Balanced Trade Reduction

5 Conclusion

8 / 28

Page 15: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Outline

1 The ModelAuction-based Ridesharing

2 Fixed-price Mechanisms

3 VCG with Reserve Prices

4 Balanced Trade Reduction

5 Conclusion

9 / 28

Page 16: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

System Overview

10 / 28

Page 17: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

System Overview

Input:Route map: a graph G = (L,E),

L: stopping points/locations,E : routes between stopping points,w(e): time required to travel via route e ∈ E .

Commuter i ’s private trip/type: θi = (ldi , lai , t

di , t

ai , ci ,qi)

ldi , lai ∈ L: departure and arrival locations,

tdi , t

ai : earliest departure and latest arrival time,

ci ∈ R+: travel cost to finish the trip,qi ∈ N: extra seats available on the trip.

10 / 28

Page 18: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

System Overview

Output:Allocation/Scheduling:

driver: drives and takes ridersrider: shares with driversunmatched: goes with the original travel preference

Payments:driver: receives moneyrider: pays moneyunmatched: no payment

10 / 28

Page 19: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

The Goal of the System

Properties of the output:Minimize the total travel costs (efficiency)Incentivize participation and against manipulations

commuters never receive negative utility (individualrationality)truthfully reporting their trip information is a dominantstrategy (truthfulness)

Deficit control (budget balance)The system owner should not lose too much money

11 / 28

Page 20: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

One Solution: Applying VCG Mechanism

One classical solution: VCG

12 / 28

Page 21: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

One Solution: Applying VCG Mechanism

Vickrey Auction (Second Price Auction)

12 / 28

Page 22: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

One Solution: Applying VCG Mechanism

Vickrey Auction (Second Price Auction)

Allocation: the agent with the highest valuation winsPayment : the harm of others caused by the winnerProperties: Efficient, Individual Rational, and Truthful.

12 / 28

Page 23: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

One Solution: Applying VCG Mechanism

Vickrey-Clarke-Groves (VCG) auction

Efficient (costs minimizing)Individually rational (agents never lose money)Truthful (truthfully reporting is the best)High deficit (m times of the costs saved!)

12 / 28

Page 24: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

One Solution: Applying VCG Mechanism

Vickrey-Clarke-Groves (VCG) auction

Efficient (costs minimizing)Individually rational (agents never lose money)Truthful (truthfully reporting is the best)High deficit (m times of the costs saved!)

12 / 28

Page 25: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

One Solution: Applying VCG Mechanism

Vickrey-Clarke-Groves (VCG) auction

Efficient (costs minimizing)Individually rational (agents never lose money)Truthful (truthfully reporting is the best)High deficit (m times of the costs saved!)

QuestionHow to control deficit?

12 / 28

Page 26: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Auction-based Ridesharing

Our Solutions (Overview)

We propose...Fixed-price Mechanisms :

Flexible deficit control (outperforms VCG)Truthful and individually rationalVery inefficient

VCG with Reserve Prices :

Flexible deficit control (outperforms VCG)(Partially) truthful and individually rationalFlexible efficiency control

Balanced Trade Reduction :

Flexible deficit control (outperforms VCG)truthful and individually rationalFlexible efficiency control

13 / 28

Page 27: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Outline

1 The Model

2 Fixed-price Mechanismsxfixed(p0,p1)

3 VCG with Reserve Prices

4 Balanced Trade Reduction

5 Conclusion

14 / 28

Page 28: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

xfixed (p0, p1)

Fixed Payments xfixed(p0,p1)

Given predefined values p0 ≥ 0 (for riders) and p1 ≤ 0 (fordrivers), fixed payments are defined

Allocation independentAllocation dependent

location dependent (e.g. shortest path)detour dependentsharing dependent...

15 / 28

Page 29: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

xfixed (p0, p1)

Dictatorship Mechanism

Serial Dictatorship Mechanism with fixed payments1 Predefine the set of (potential) drivers and riders2 Order potential drivers and riders3 Maximize drivers’ utility according to the order4 Each driver/rider gets the fixed payment

Propertiestruthful and individually rationalbetter deficit control than VCGvery inefficient

16 / 28

Page 30: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

xfixed (p0, p1)

Problems of Non-dictatorship Mechanisms

Case I:fixedPay = 10both prefer drivingpotential problem fordeterministic mechanisms

Case II:fixedPay = 1both prefer ridingpotential problem for allmechanisms

17 / 28

Page 31: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Outline

1 The Model

2 Fixed-price Mechanisms

3 VCG with Reserve PricesMVCG(r0, r1)

4 Balanced Trade Reduction

5 Conclusion

18 / 28

Page 32: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

MVCG(r0, r1)

VCG with Two-Sided Reserve PricesMVCG(r0, r1)

Predefined reserve prices r0 ≥ 0 (for riders) and r1 ≤ 0 (fordrivers),

Note: r0 and r1 can be allocation dependent.

19 / 28

Page 33: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

MVCG(r0, r1)

Properties ofMVCG(r0, r1)

MVCG(r0, r1) istruthful iff r0 ≥ −r1. Otherwise, the manipulationgain is bounded (max(−r1 − r0, δmax

i (−r1 − r0))).weakly budget balanced without detour.Otherwise, deficit is bounded (−ndδ

max r1 − nr r0).more efficient as r0 + r1 decreases.

20 / 28

Page 34: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

MVCG(r0, r1)

A Common Feature of the Dictatorship and VCG withReserves

The truthfulness is satisfied ifeach commuter can only be allocated as one predefinedrole (driver or rider) or unmatched.

21 / 28

Page 35: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Outline

1 The Model

2 Fixed-price Mechanisms

3 VCG with Reserve Prices

4 Balanced Trade ReductionMcAfee’s Trade ReductionBalanced Trade Reduction

5 Conclusion

22 / 28

Page 36: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

McAfee’s Trade Reduction

McAfee’s Trade Reduction (1992)

McAfee’s reduction: increase VCG payments via reducingefficiency

94

98

113

90

85

79

53

121

82

91

112

82

69

37

120

125

sellers/

drivers

buyers/

riders

VCG payments:Riders: 90Drivers: −91

Deficit: 4McAfee’s payments:

Riders: 91Drivers: −90

No deficit!

23 / 28

Page 37: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

McAfee’s Trade Reduction

McAfee’s Reduction is NOT Truthful in Ridesharing

McAfee’s trade reduction is NOT truthful in ridesharingbecause:

a commuter who can be allocated as either driver or ridermight manipulate/switch!

24 / 28

Page 38: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Balanced Trade Reduction

Balanced Trade Reduction (BTR)

VCG payment for a buyer

vbi − xvcg

i ≥ vsi − x̂vcg

i

McAfee’s payment for a buyer

vbi − (xvcg

i + δ1) � vsi − (x̂vcg

i + δ2)

Balanced Trade Reduction payment for a buyer

vbi − (xvcg

i + δ) ≥ vsi − (x̂vcg

i + δ)

25 / 28

Page 39: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Balanced Trade Reduction

Balanced Trade Reduction (BTR)

VCG payment for a buyer

vbi − xvcg

i ≥ vsi − x̂vcg

i

McAfee’s payment for a buyer

vbi − (xvcg

i + δ1) � vsi − (x̂vcg

i + δ2)

Balanced Trade Reduction payment for a buyer

vbi − (xvcg

i + δ) ≥ vsi − (x̂vcg

i + δ)

Properties of BTR:Truthful, Individual Rational, almost Efficient, but hasDeficit.

25 / 28

Page 40: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Outline

1 The Model

2 Fixed-price Mechanisms

3 VCG with Reserve Prices

4 Balanced Trade Reduction

5 ConclusionDone and ToDo

26 / 28

Page 41: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Done and ToDo

What is NEW?

The first comprehensive ridesharing model studied from apure game-theoretic point of view.Auction-based ridesharing system incentivizingparticipation.Flexible deficit control rather than completely removedeficit.

27 / 28

Page 42: SMART Seminar Series: Mechanism Design for Ridesharing

The Model Fixed-price Mechanisms VCG with Reserve Prices Balanced Trade Reduction Conclusion

Done and ToDo

Future Work

Tradeoff between deficit and efficiency (theoretically orsimulations).The problem of finding optimal schedules iscomputationally hard (optimal in range).Allow agents to submit trips dynamically over time (onlinemechanism design).

28 / 28


Recommended