Sustainable Passenger Transportation: Dynamic Ride-sharing
Martin Savelsbergh
Joint work with Niels Agatz, Alan Erera, Xing Wang TRANSLOG December 9, Chile
Motivation
Observations
Congestion is a major issue in urban areas around the world Congestion leads to
Loss of productivity Wasted fuel Pollution
Private car occupancies are low
Average of 1.8 for leisure trips Average of 1.1 for commuting trips
Motivation
Opportunity
Effective use of empty car seats can
Reduce congestion Reduce fuel consumption Reduce pollution Increase productivity
Dynamic Ride-Sharing
Enabling technology exists
GPS-enabled mobile devices Mobile-to-mobile communication Route guidance technology
Outline
Introduction to dynamic ride-sharing Comparison with on-demand transportation Dynamic ride-sharing variants Multi-modal transit systems Key challenges Research opportunities
Dynamic ride-sharing: basics
1. 2. 3. 4.
Announcement of trips Matching of drivers and riders Execution of identified trips Sharing of trip costs
Assumption: if no trip is identified, drivers and riders use their own car to reach destination
Dynamic ride-sharing: features
Dynamic: established on short-notice Non-recurring trips: different from carpooling Automated matching: different from online noticeboards Independent drivers: different from taxis Cost-sharing: variable trip related expenses Prearranged: different from spontaneous, casual ride-sharing
Dynamic ride-sharing: providers
IPhone applications:
Carticipate (Aug. 08) Avego (Dec. 08)
Google Android applications:
Piggyback (May 08)
Web-based:
Zebigo, PickupPal, Nuride, Rideshark, Ridecell, Goloco, Zimride, , etc.
What is in it for participants?
Save costs by sharing fuel expenses and tolls Save time through use of HOV-lanes Save the planet
What is in it for providers?
Private: take a cut of the participants cost savings Public (society): decrease external costs of transportation: emissions, pollution, and traffic congestion
Objectives
Participants: travel costs Private ride-share provider: total travel costs cut Society: pollution and traffic congestion Objectives are aligned: system-wide vehicle miles, win-win-win
Time is crucial
Time is likely to be more constraining than out-of-distance or the number of spare seats Information exchange: departure time alone may not suffice
departure time flexibility? travel time flexibility?
Time model
time window for matching flexibility direct travel time
earliest departure time announcement time
latest departure time
latest arrival time
Inter-personal constraints
smoking non-smoking female male friends strangers subsets of feasible ride-share partners? integration with social networking tools!
How to divide the trip costs?
Allocate trip costs (savings) proportional to the original trip costs? Financial aspects handled by provider
Related problems/literature
Unscheduled, on-demand passenger transportation
Dial-a-ride Taxi Mobility Allowance Shuttle Transport
Carpooling Scheduled passenger transporation
Comparison with on-demand passenger transportation
Product: door-to-door transportation Capacity: company vehicles independent private drivers Revenues: ticket price cost-sharing Route: round-trip from depot one way from drivers origin to destination Dynamics: new requests new riders and new drivers
Dynamic ride-share variants
Single rider Bipartite Matching = easy Multiple riders
Single driver Multiple drivers
Illustrative example: Single driver single rider
3 drivers 2 riders Earliest departure / latest arrival time Travel time ~ travel distance Miles = minutes (60mph)
Announced trips
[7:50] d1 8 [8:11] d1
2.8 [7:55] r1 5 [7:58] d2 1.6 2.1 6 5 2.1 1.6 4 [8:04] r1
2.8
5 [8:15] d2
2 1.8 r2 [8:02]
d3 [8:00]
d3 [8:12] 1.8
r2 [8:18]
Incompatible ride-share pair: d3-r1
[7:50] d1 8 [8:11] d1
2.8 [7:55] r1 5 [7:58] d2 1.6 2.1
Time infeasible!
4
2.8 [8:04] r1 5 1.6 2.1 [8:15] d2
6 5
2 1.8 r2 [8:02]
d3 [8:00]
d3 [8:12] 1.8
r2 [8:18]
Incompatible ride-share pair: d1-r2
[7:50] d1 8 [8:11] d1
2.8 [7:55] r1 5 [7:58] d2 1.6 2.1
o savings!
[8:04] 4 r1
2.8
5 1.6 2.1 6 5 d2 [8:15]
2 1.8 r2 [8:02]
d3 [8:00]
d3 [8:12] 1.8
r2 [8:18]
Bipartite matching
d1 2.4 r1
2.8
weight = cost savings
r2 Riders
d2
1.4
d3 Drivers
Bipartite matching
d1 2.4 r1
2.8
weight = cost savings
r2
Riders
d2
Solution: d1-r1
1.4
d3
Drivers
d2-r2
Optimal solution
[7:50] d1 [8:11]
2.8
17% savings in system-wide vehicle miles!
[7:55] r1 4 [8:04] r1
d1
2.8
[7:58] d2 5 d2
[8:15]
d3 [8:00]
d3 [8:12]
r2 [8:02]
r2 [8:18]
System versus user benefits
d1 2.4 r1
2.8
d2
r2
Riders
1.4
d3
Drivers
User benefits
[7:50] d1 [8:11] d1
[7:55] r1
[8:04] r1
[7:58] d2 d2
[8:15]
d3 [8:00] r2 [8:02]
d3 [8:12] r2 [8:18]
Dynamic ride-share variants
Single rider Easy (Bipartite Matching) Difficult (Transfers) Multiple riders Difficult (Routing) Difficult (Routing + Transfers)
Single driver Multiple drivers
Multi-modal transit systems
Ride-sharing as feeder for public transport e.g. first and last mile to and from station Taxis as backup option for ride-sharing? Multi-modal system design
What are the key challenges to implementation?
1. Reaching a critical mass
How to ensure participation in the start-up phase when service is low? What kind of incentives can be offered? What role should (local) government and business communities play? What role can/should taxis play?
2. Effectively handling the dynamics
New drivers and riders continuously enter and leave the system! When to commit to a ride-share? What response time is acceptable for users? Should en-route matching be considered?
r2
3. Ensuring safety and reliability
Reputation systems: let ride-share partners rate each other! Reward good ratings! Use GPS-technology to monitor trip behavior Back-up options / Return trips
3. Ensuring safety and reliability
Use social network tools to match up friends and friends-of-friends What kind of social networks are conducive to ride-sharing?
shape of social network position in network, number of friends overlap between professional and private networks geographic density
How can the transportationoptimization community contribute?
Basic research opportunities
Optimization
Fast matching algorithms
Single rider, single driver Multiple riders per trip (routing) Multiple drivers per trip (transfers)
Handling of system dynamics
Rolling horizon framework
Incentive schemes Multi-modal settings (ride-sharing & public transit)
Simulation
Create an environment to study impact of system characteristics Create an environment to advice policy makers
Under Development
An instance generator A simulation environment based on traffic model data from Atlanta Regional Council Matching technology for single rider single driver setting
Dynamic Ride-Sharing Simulation of Metro Atlanta
Metro Atlanta
population: 4.70 million area: 4 million acres (16,000 km2)
Traffic Model Atlanta Regional Council
2024 Traffic Analysis Zones (TAZs) # work-related round trips per day: 2.96 million ~ 60% single occupancy trips # O-D pairs: 2.90 million max # trips per O-D pair: 881 min # trips per O-D pair: 0.01
Future research opportunities
More realistic travel time settings
Time-dependent travel times Real-time travel times
Behavioral modeling
Match acceptance Cancelations No-shows
Variability
Late driver departure Late rider departure
Future research opportunities (cont.)
Optimization
Centralized vs. decentralized matching Buffering vs. no buffering Anticipating future announcements
Thank You
