Inferring Unmet Demand from Taxi Probe Data
ITSC 2015
Inferring Unmet Demand from Taxi Probe Data Afian Anwar (MIT), Amedeo Odoni (MIT), Daniela Rus (MIT)
ITSC 2015
MOTIVATION
2
ITSC 2015
TAXI BOOKING SYSTEMS ARE INADEQUATE Traditional booking systems match passengers to drivers within a fixed search radius.
3
ITSC 2015
TAXI BOOKING SYSTEMS ARE INADEQUATE Traditional booking systems match passengers to drivers within a fixed search radius. These systems fail completely when there is a spatial-temporal mismatch in supply and demand.
4
ITSC 2015
TAXI BOOKING SYSTEMS ARE INADEQUATE Traditional booking systems match passengers to drivers within a fixed search radius. These systems fail completely when there is a spatial-temporal mismatch in supply and demand. We provide a simple, scalable way to use taxi probe data to measure unmet taxi demand in real time.
5
ITSC 2015
TAXI BOOKING SYSTEMS ARE INADEQUATE Traditional booking systems match passengers to drivers within a fixed search radius. These systems fail completely when there is a spatial-temporal mismatch in supply and demand. Our work aims to solve this problem by providing drivers with real time unmet demand information in an app.
6
ITSC 2015
PRIOR ART
http://therideshareguy.com/lyfts-heat-maps-vs-ubers-surge-pricing-who-wins/ 7
ITSC 2015
PRIOR ART Modern ride sharing information systems rely on surge pricing to provide information. Reactive, not predictive. Does not apply to non-surge pricing business models e.g. traditional taxi fleets.
http://therideshareguy.com/lyfts-heat-maps-vs-ubers-surge-pricing-who-wins/ 8
ITSC 2015
PRIOR ART Significant body of academic research to develop data driven decision support tools that enable taxi drivers to operate more efficiently. A novel approach to independent taxi scheduling problem based on stable matching R. Bai, J. Li, J. A. Atkin, and G. Kendall Analysis of the passenger pick-up pattern for taxi location recommendation J. Lee, I. Shin, and G.-L. Park Dynamic Patrolling Policy for Optimizing Urban Mobility Networks (ITSC 2015) G. Hall, M. Volkov and D. Rus
9
ITSC 2015
PROBLEM FORMULATION We measure unmet demand by the quantity U, which is the answer to the question: “How many more taxis are needed in an area to completely satisfy all taxi demand for a given period of time?” D = 8 (passenger demand D)
10
ITSC 2015
PROBLEM FORMULATION We measure unmet demand by the quantity U, which is the answer to the question: “How many more taxis are needed in an area to completely satisfy all taxi demand for a given period of time?” D = 8 (passenger demand D)
S = B = 2 (taxi supply S = taxi boardings B)
11
ITSC 2015
PROBLEM FORMULATION We measure unmet demand by the quantity U, which is the answer to the question: “How many more taxis are needed in an area to completely satisfy all taxi demand for a given period of time?” D = 8 (passenger demand D) U=D-B =D-S =8-2=6 S = B = 2 (taxi supply)
12
ITSC 2015
PROBLEM FORMULATION We measure unmet demand by the quantity U, which is the answer to the question: “How many more taxis are needed in an area to completely satisfy all taxi demand for a given period of time?” D = 8 (passenger demand D) U=D-B =D-S =8-2=6 S = B = 2 (taxi supply)
13
ITSC 2015
PROBLEM FORMULATION We develop an indicator, the unmet demand intensity p which is positively correlated to U. For the special case when taxi supply is in excess i.e. S - B > 0, D = B we show that p=k
(ratio of demand to residual taxi queue length) D = 2 (passenger demand D)
S = 5 (taxi supply)
14
ITSC 2015
PROBLEM FORMULATION We develop an indicator, the unmet demand intensity p which is positively correlated to U. For the special case when taxi supply is in excess i.e. S - B > 0, D = B we show that p=k =
(ratio of demand to residual taxi queue length) D = 2 (passenger demand D)
S = 5 (taxi supply)
15
ITSC 2015
PROBLEM FORMULATION We develop an indicator, the unmet demand intensity p which is positively correlated to U. For the special case when taxi supply is in excess i.e. S - B > 0, D = B we show that p=k =
(ratio of demand to residual taxi queue length) D = 2 (passenger demand D)
S = 5 (taxi supply) B = 2 (matched demand = taxi boardings) S - B = 3 (residual queue length of taxis) 16
ITSC 2015
ASSUMPTIONS 1. Exactly one passenger boards a taxi 2. Taxi status can only take values FREE or POB 3. Time is discretized - data received at exact intervals 4. Demand / supply clears completely within the observed time 5. More taxis than people
17
ITSC 2015
DATA
{loc, taxi_id, time, status}
18
ITSC 2015
PROBLEM FORMULATION We develop an indicator, the unmet demand intensity p which is positively correlated to U. For the special case when taxi supply is in excess i.e. S - B > 0, D = B we show that p=k
(ratio of demand to residual taxi queue length)
=
19
ITSC 2015
DATA ANALYSIS
20
ITSC 2015
DATA ANALYSIS
21
ITSC 2015
TAXI SLACK
Qn: When was taxi supply greatest?
22
ITSC 2015
TAXI SLACK
Qn: When was taxi supply greatest? Ans: Time period 3
23
ITSC 2015
TAXI SLACK
24
ITSC 2015
TAXI SLACK
25
ITSC 2015
TAXI BOARDINGS
26
ITSC 2015
UNMET DEMAND INTENSITY
1
1 2
2
27
ITSC 2015
PROOF SKETCH To show: p=k
28
ITSC 2015
PROOF SKETCH To show: p=k From previous slide:
29
ITSC 2015
PROOF SKETCH To show: p=k From previous slide:
30
ITSC 2015
ANALYSIS
D=
D = 2 (passenger arrival rate)
D: taxi demand S: taxi supply B: taxi boardings 31
ITSC 2015
ANALYSIS
D= B=
D = 2 (passenger arrival rate) D = 2 (passenger boarding rate)
B=D
D: taxi demand S: taxi supply B: taxi boardings 32
ITSC 2015
PROOF SKETCH To show: p=k From previous slide:
=
33
ITSC 2015
ANALYSIS
taxis
time 34
ITSC 2015
ANALYSIS
taxis
time 35
ITSC 2015
TAXI DEMAND >> TAXI SUPPLY
36
ITSC 2015
EXPERIMENTAL RESULTS
Harborfront / Vivocity (18:00 hrs)
37
ITSC 2015
UNMET DEMAND APP
38
ITSC 2015
DEPLOYMENT TO AUTONOMOUS VEHICLE FLEET
http://smart.mit.edu 39
Q&A
A big thank you to my MIT colleagues Mikhail Volkov and Gavin Hall and advisors Amedeo Odoni and Daniela Rus 40
Inferring Unmet Demand from Taxi Probe Data Afian Anwar (MIT), Amedeo Odoni (MIT), Daniela Rus (MIT)
ITSC 2015
MOTIVATION
2
ITSC 2015
TAXI BOOKING SYSTEMS ARE INADEQUATE Traditional booking systems match passengers to drivers within a fixed search radius.
3
ITSC 2015
TAXI BOOKING SYSTEMS ARE INADEQUATE Traditional booking systems match passengers to drivers within a fixed search radius. These systems fail completely when there is a spatial-temporal mismatch in supply and demand.
4
ITSC 2015
TAXI BOOKING SYSTEMS ARE INADEQUATE Traditional booking systems match passengers to drivers within a fixed search radius. These systems fail completely when there is a spatial-temporal mismatch in supply and demand. We provide a simple, scalable way to use taxi probe data to measure unmet taxi demand in real time.
5
ITSC 2015
TAXI BOOKING SYSTEMS ARE INADEQUATE Traditional booking systems match passengers to drivers within a fixed search radius. These systems fail completely when there is a spatial-temporal mismatch in supply and demand. Our work aims to solve this problem by providing drivers with real time unmet demand information in an app.
6
ITSC 2015
PRIOR ART
http://therideshareguy.com/lyfts-heat-maps-vs-ubers-surge-pricing-who-wins/ 7
ITSC 2015
PRIOR ART Modern ride sharing information systems rely on surge pricing to provide information. Reactive, not predictive. Does not apply to non-surge pricing business models e.g. traditional taxi fleets.
http://therideshareguy.com/lyfts-heat-maps-vs-ubers-surge-pricing-who-wins/ 8
ITSC 2015
PRIOR ART Significant body of academic research to develop data driven decision support tools that enable taxi drivers to operate more efficiently. A novel approach to independent taxi scheduling problem based on stable matching R. Bai, J. Li, J. A. Atkin, and G. Kendall Analysis of the passenger pick-up pattern for taxi location recommendation J. Lee, I. Shin, and G.-L. Park Dynamic Patrolling Policy for Optimizing Urban Mobility Networks (ITSC 2015) G. Hall, M. Volkov and D. Rus
9
ITSC 2015
PROBLEM FORMULATION We measure unmet demand by the quantity U, which is the answer to the question: “How many more taxis are needed in an area to completely satisfy all taxi demand for a given period of time?” D = 8 (passenger demand D)
10
ITSC 2015
PROBLEM FORMULATION We measure unmet demand by the quantity U, which is the answer to the question: “How many more taxis are needed in an area to completely satisfy all taxi demand for a given period of time?” D = 8 (passenger demand D)
S = B = 2 (taxi supply S = taxi boardings B)
11
ITSC 2015
PROBLEM FORMULATION We measure unmet demand by the quantity U, which is the answer to the question: “How many more taxis are needed in an area to completely satisfy all taxi demand for a given period of time?” D = 8 (passenger demand D) U=D-B =D-S =8-2=6 S = B = 2 (taxi supply)
12
ITSC 2015
PROBLEM FORMULATION We measure unmet demand by the quantity U, which is the answer to the question: “How many more taxis are needed in an area to completely satisfy all taxi demand for a given period of time?” D = 8 (passenger demand D) U=D-B =D-S =8-2=6 S = B = 2 (taxi supply)
13
ITSC 2015
PROBLEM FORMULATION We develop an indicator, the unmet demand intensity p which is positively correlated to U. For the special case when taxi supply is in excess i.e. S - B > 0, D = B we show that p=k
(ratio of demand to residual taxi queue length) D = 2 (passenger demand D)
S = 5 (taxi supply)
14
ITSC 2015
PROBLEM FORMULATION We develop an indicator, the unmet demand intensity p which is positively correlated to U. For the special case when taxi supply is in excess i.e. S - B > 0, D = B we show that p=k =
(ratio of demand to residual taxi queue length) D = 2 (passenger demand D)
S = 5 (taxi supply)
15
ITSC 2015
PROBLEM FORMULATION We develop an indicator, the unmet demand intensity p which is positively correlated to U. For the special case when taxi supply is in excess i.e. S - B > 0, D = B we show that p=k =
(ratio of demand to residual taxi queue length) D = 2 (passenger demand D)
S = 5 (taxi supply) B = 2 (matched demand = taxi boardings) S - B = 3 (residual queue length of taxis) 16
ITSC 2015
ASSUMPTIONS 1. Exactly one passenger boards a taxi 2. Taxi status can only take values FREE or POB 3. Time is discretized - data received at exact intervals 4. Demand / supply clears completely within the observed time 5. More taxis than people
17
ITSC 2015
DATA
{loc, taxi_id, time, status}
18
ITSC 2015
PROBLEM FORMULATION We develop an indicator, the unmet demand intensity p which is positively correlated to U. For the special case when taxi supply is in excess i.e. S - B > 0, D = B we show that p=k
(ratio of demand to residual taxi queue length)
=
19
ITSC 2015
DATA ANALYSIS
20
ITSC 2015
DATA ANALYSIS
21
ITSC 2015
TAXI SLACK
Qn: When was taxi supply greatest?
22
ITSC 2015
TAXI SLACK
Qn: When was taxi supply greatest? Ans: Time period 3
23
ITSC 2015
TAXI SLACK
24
ITSC 2015
TAXI SLACK
25
ITSC 2015
TAXI BOARDINGS
26
ITSC 2015
UNMET DEMAND INTENSITY
1
1 2
2
27
ITSC 2015
PROOF SKETCH To show: p=k
28
ITSC 2015
PROOF SKETCH To show: p=k From previous slide:
29
ITSC 2015
PROOF SKETCH To show: p=k From previous slide:
30
ITSC 2015
ANALYSIS
D=
D = 2 (passenger arrival rate)
D: taxi demand S: taxi supply B: taxi boardings 31
ITSC 2015
ANALYSIS
D= B=
D = 2 (passenger arrival rate) D = 2 (passenger boarding rate)
B=D
D: taxi demand S: taxi supply B: taxi boardings 32
ITSC 2015
PROOF SKETCH To show: p=k From previous slide:
=
33
ITSC 2015
ANALYSIS
taxis
time 34
ITSC 2015
ANALYSIS
taxis
time 35
ITSC 2015
TAXI DEMAND >> TAXI SUPPLY
36
ITSC 2015
EXPERIMENTAL RESULTS
Harborfront / Vivocity (18:00 hrs)
37
ITSC 2015
UNMET DEMAND APP
38
ITSC 2015
DEPLOYMENT TO AUTONOMOUS VEHICLE FLEET
http://smart.mit.edu 39
Q&A
A big thank you to my MIT colleagues Mikhail Volkov and Gavin Hall and advisors Amedeo Odoni and Daniela Rus 40
