D-STOP Symposium 2016Closing Remarks

Sanjay ShakkottaiDirector, Wireless Networking and Communications Group

1

Wireless Networking and Communications Group

125 Grad Students

Affiliates champion large federal proposals, provide technical input/feedback, research support

WNCG provides pre-competitive research, technical expertise, first access to students

Significant number of students intern/work full-time for affiliates

Affiliates provide real world context

Industrial Affiliates22 Faculty

2007-08

2009-10

2011-12

2013-14 $-

$2,000,000.00 $4,000,000.00 $6,000,000.00

Data Enabled Multi-scale Platform for Planning and Operations

Models and Learning Algorithms1. New regression algorithms for

noisy and high-dimensional data (leverage vast offline data)

2. Online learning for real-time decisions (leverage real-time data)

3. Models for high dimensional dataServices and Applications

1. Adaptive traffic signaling2. Transit planning using network models

The Internet of Moving Things1. Vehicles as dynamic sensors2. Agile V2X communications

Image source: http://smartdesignworldwide.com ; Full link at: http://bit.ly/1MzYxMe

Data Portal & Analytics

1. Hosted at UT’S TACC2. TMC testbed3. Versioning capabilities4. Edit via free tools or ArcGIS

Public release of select data

Community

1. OpenSteetMap2. City and State

open data portalsPublished and crowd-sourced data

Transportation Agencies

Private Sector

1. App Developers2. OEMs

Agen

cy

Dat

a

Regional D

ata

Select

Access

Data Data

Open

Acce

ss

The Data Analytics Platform

Additional details: Please contact Dr. Jennifer DuthieMap image screenshot from http://www.openstreetmap.org , Transport Management Center: Source: http://www.moxa.com ; Full link at http://bit.ly/1qfBJrd

Vehicles as

Sensors

Data

The Internet of Moving Things

Sensing + Communications

Infrastructure-based sensingSensing includes radar,

LIDAR, cameras, and weather

Coordinate traffic through intersections, support

automated driving

Collect data about collisions and near-misses for

planningEffective with non-connected

cars, bicycles, and pedestrians

Sensing includes radar, LIDAR, cameras, and

weather

Coordinate traffic through intersections, support

automated driving

Collect data about collisions and near-misses for

planningEffective with non-connected

cars, bicycles, and pedestrians

Radar-aided millimeter wave communication

mmWave BS supporting V2X+radar

antennas

Radar beam

Millimeter wave is used for both radar sensing and high bandwidth communication

communication beams

Radar can be used to configure communication link more

efficiently

Additional details: Please contact Prof. Robert Heath

Dual low-cost (~$5) GPS/GNSSantennas mounted on vehicle

Standard GPS/GNSS positioning exhibits2-3-meter errors (actual traces)

Precise GPS/GNSS positioning exhibits2-3-centimeter errors (mockup trace—systemnot yet operational)

Carrier-phase-based processing of GPS/GNSS signals enables 100x improvement in accuracy compared to standard GPS/GNSS

positioning. But current costs are too high (~$2k). WNCG researchers are developing a ~$50 sensor that achieves reliable instantaneous

precise positioning.

Applications

Lane violation statistics: Where are drivers routinely departing the lane?

Lane-responsive signaling

Intuitive heads-up-display: Driver sees path to destination “painted” on roadway

“Last moment” lane keeping: Vehicle nudges car back into lane only when unintentional lane departure is imminent. Unlike Tesla’s Autopilot, this keeps driver engaged

Densely-space reference stations compensate for GPS/GNSS signal atmospheric delays so that vehicles can be instantaneously positioned to sub-decimeter globally-referenced accuracy. UT-Samsung centimeter-accurate mobile positioning system (CAMPS) reference network in Austin,

Texas, with site hosting courtesy of TxDOT. Additional details: Please contact Prof. Todd Humphreys

Communications – The Internet of Moving Things (IoMT)

Full Duplex Radios - Can transmit & receive at the same frequency at same time Thought to be impossible 8 years ago

Possible through self-interference isolation and cancellation 110+dB of isolation/cancellation

necessary Enables listen-while-talking

Much more efficient mobile meshing Full duplex + mobile meshing

Low overhead, high throughput meshing

Connect people and vehicles as they move Discovery, routing, handoff efficient

Additional details: Please contact Prof. Sriram Vishwanath

Algorithms for Large Scale Learning

Offline + Online Learning

Mixtures and Non-Linearities in Large Scale Data Analysis

Linear, Logistic and Non-linear regression are fundamental for prediction and planning Examples: transit time vs. daily flows, flow vs. speed, responses to network

stressors or diversions or to future demand and flow patterns Mixtures: Populations are mixed, and may require simultaneous clustering and

regression/classification, when clustering-as-data-preprocessing is impossible Nonlinearities: Discover structure without expensive/intractable non-parametric

models

New algorithms for:

1. Solving the simultaneous clustering-regression problem (tensor methods)

2. Structure recovery through unknown non-linear transforms (second-moment methods)

Additional details: Please contact Prof. Constantine Caramanis

45 th St

38 ½ th St

I-35

Red R

iver S

t

Airport Blvd

Guad

alupe

St

Lam

ar B

lvd

32 nd St

Manor Rd

51 st St

Northfield

Windsor Park

RidgetopHyde Park/ Northfield

Delwood II

Hancock

North University

Cherrywood/Wilshire Wood / Delwood I

MuellerBarbara Jordan Blvd

38 ½th St

Manor Rd

Other

Ramps used by neighborhood traffic, Source: Dr. J. Duthie

Online Decision Making – Bandit Algorithms Online learning algorithms for real-time matching between servers and demand

Freight: Servers == trucks; Demand == packages/cargo Travel: Servers == cars; Demand == passengers

Server availability and demand varies with time Service time is random Market matching only if servers and demand available

New algorithms based on queueing bandits for online learning and resource allocation

Additional details: Please contact Prof. Sanjay Shakkottai

Source: http://volvotrucks.com ; Full link at: http://bit.ly/1WWobLD

Modeling High Dimensional Heterogeneous Data

New Spatial Generalized Heterogeneous Data Model

Correlation across various dimensions (of the dependent variables) are captured using latent constructs

Maximum Approximate Composite Marginal Likelihood (MACML) estimation approach is used for estimation of GHDM

Source: http://www.networkworld.com ; Full link at: http://bit.ly/235TMOy

Additional details: Please contact Prof. Chandra Bhat

Multimodal data: conventional sources + cameras, GPS, cell phone tracking Methodologies to combine and aggregate high dimensional heterogeneous data

Applications

Real-time Operations and Long-term Planning

Pre-timed: Static signal timings cannot respond to real-time conditions

Create progression and synchronize operation for maximum flow under a deterministic load

Actuation: Intersections use sensors to detect waiting vehicles and adjust signal timings responsively

Can respond to current demand, but lose benefits of coordination

Both paradigms have significant drawbacks that can be overcome with new technology

Source: http://www.moxa.com ; Full link at http://bit.ly/1qfBJrd

Real-time Data-driven Signal Timing

Additional details: Please contact Profs. Boyles / Shakkottai

New algorithms seek the best of both paradigms: adaptive control with global optimality properties Sensors can measure queue lengths and estimate turn fractions and (approximate) destinations Analogous to packet routing problems in telecommunication networks, where fast, decentralized algorithms exist

Source: http://www.moxa.com ; Full link at http://bit.ly/1qfBJrd

Application: Data-driven Signal Timing

Source: US DOT ; Full link at http://1.usa.gov/1pIKhGv

Bus Transit Planning along the Guadalupe Corridor

Estimates of Guadalupe corridor boardings and alightings

Data-driven modeling and planning to study various “what-if” scenarios:1. Dedicate one lane to buses2. Move buses to parallel

corridor3. Transit-only lane + queue

jump and signal priority

Additional details: Please contact Dr. Jennifer Duthie

Conclusion Moving towards an integrated platform spanning sensing,

algorithms and applicationsGoals are to support both real-time operations and long-term

planning

D-STOP Center cross-cutting research spans multiple disciplinesCollaborations across disciplines to develop new methods and

algorithms