Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of...

23

Transcript of Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of...

Page 1: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private
Page 2: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Theja TulabandhulaXerox Research Centre India

Optimal Automated Booking of On-Demand Transportation

Page 3: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Urban Mobility at Xerox

Providers

Mobility Marketplace

Private sector fragmented,

un-coordinated, confusing

1. “I want to go

from here to B.”

2. App suggests public

and private options.

3. Private providers pay fee on rides referred

to them, regardless of payment processing.

Page 4: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Urban Mobility at Xerox

• GO-LA: Launched in Los Angeles on Jan 27, 2016

• “Xerox Built the Ultimate Transportation App for Los Angeles” - Forbes

Page 5: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

• MMTPs suggest trip plans that involve private transportation modes on some legs.

• Commuters may have to book these services by themselves

• This is inefficient. Should be automated.

• Some MMTPs may attempt to automate these bookings. But they will be still very

suboptimal.

SUBOPTIMALITY IS DUE TO TRANSFER POINT DELAYS.

*MMTP: Multi-Modal Trip Planner

Motivation

Page 6: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Motivation

• Consequences:

• Commuter has to wait: bad experience, increased travel time

• Service has to wait: more cost to the commuter

SUBOPTIMALITY IS DUE TO TRANSFER POINT DELAYS.

Page 7: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Our Contribution

Problem: How to optimally and automatically book on-demand transportation

services in multi-modal trips to help commuters be cost-efficient?

No solution exists that optimally and automatically books on-demand

transportation services.

Page 8: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Solution: Optimal Automated Booking (OAB)

• OAB handles optimal booking of private transportation modes to minimize waiting time costs at transfer points.

• Uses ETA of current leg of the journey to destination (NextBus etc)

• Uses forecasted availabilities of services (Uber API etc)

• Automatically performs booking and sends confirmation to user

Page 9: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Activity Diagram

Page 10: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Layered System Architecture

Page 11: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

OAB (Optimal Automated Booking) Strategy

• Discrete time indexed by t

• Four quantities of interest:

– T: Estimated Time of Arrival (ETA) of the current mode

– S(t): Service Availability Time of the on-demand mode

– h: Per unit commuter waiting cost > 0

– b: Per unit on-demand service waiting cost for the commuter > 0

• Intuition: If T > S(t), then the service waits and charges commuter $b

per unit of time. If T < S(t), then the commuter waits and this costs

her/him $h per unit of time.

Note: t, T and S(t) are absolute times

Page 12: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

How does OAB work? Setting I

Stochastic model for ETA and deterministic model for Service Availability is fixed and known

Analytical formula for

the optimal booking time

Page 13: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

How does OAB work? Setting II

Stochastic models for ETA and Service Availability are fixed and known

Analytical formula for

the optimal booking time

Page 14: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

How does OAB work? Setting IIIStochastic models for ETA and Service Availability are not known beforehand and are time

varying

Page 15: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Experiment: Setting I

Arrival Model: Uni-modal distribution (green) with mean ETA 200 minutes and standard deviation 20 minutes

Service Availability Model: Always available with deterministic delay of 5 minutes

User waiting cost is same as service waiting cost

Page 16: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Experiment: Setting I

Arrival Model: Uni-modal distribution (green) with mean ETA 200 minutes and standard deviation 20 minutes

Service Availability Model: Always available with deterministic delay of 5 minutes

User waiting cost is half the service waiting cost

Page 17: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Experiment: Setting I

Arrival Model: Uni-modal distribution (green) with mean ETA 200 minutes and standard deviation 20 minutes

Service Availability Model: Always available with deterministic delay of 5 minutes

User waiting cost is double the service waiting cost

Page 18: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Experiment: Setting III (time varying data)Arrival Model: Uni-modal distribution (green) with mean ETA 200 minutes and standard deviation 20 minutes

Service Availability Model: Stochastic with mean delay of 5 minutes and standard deviation 5 minutes

User waiting cost is same as service waiting cost

Page 19: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Experiment: Setting III (time varying data)

User waiting cost is half the service waiting cost User waiting cost is double the service waiting cost

Arrival Model: Uni-modal distribution (green) with mean ETA 200 minutes and standard deviation 20 minutes

Service Availability Model: Stochastic with mean delay of 5 minutes and standard deviation 5 minutes

Page 20: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Experiment: Setting III (Boston Data)Arrival Model: From NextBus

Service Availability Model: From Uber

Note: User waiting cost is same as service waiting cost

Page 21: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Benefit for Commuters: Single-Source App

• Convenient single-source for all public and private mobility options

• Personalized to user’s preference (e.g., health, environment, cost, time)

• Unified account-based payments

• Trusted official mobility app of city

• Dynamic re-routing options

Point-to-point trip planning:

bridging first and last mile

A B

A B

parking

A B

shared van bike share

car share

Page 22: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Optimization across all mobility activity for load balancing and reduced congestion:

• Nudging people to use public transportation, ride sharing

• Automated demand-based service provision (e.g. shared shuttle)

• A city dashboard provides real-time insight for capacity planning

Add a Bike Share Station here

A B

Real-time Automated Decisions Decision support for City Planning

Dynamic Ride Request Pooling

shared van

A BA BA B

AAAA

BBBB

Increase Bus Frequency

A B

A B

Benefit for City: Reduced Congestion and System

Optimization

Page 23: Optimal Automated Booking of On-Demand Transportation€¦ · Optimal Automated Booking of On-Demand Transportation. Urban Mobility at Xerox Providers Mobility Marketplace Private

Summary

• Multi-modal journeys have travel point delays

• We have proposed online algorithms that book on-demand modes

based on predicted delays

• This saves “costs” and minimizes inconvenience for the user

• Is a building block towards more convenient multi-modal trip

experiences.

Please refer to the paper for variants and extensions of the modeling setting.