Toward a new generation of traffic control systems Marco Zennaro Prof. Raja Sengupta

Toward a new generation of traffic control systems

Marco Zennaro

Prof. Raja Sengupta

Outline

• Current traffic control systems development process;

• A new development process;• Case study;• Wireless communication, distributed

computing and traffic control systems;• Distribution problems;• Proposed solution;• Approach advantages;

Today’s traffic controllers

Multiple control devices are currently in use;

2070 controller:– General purpose computers:

• Eagle 2001 (Motorola 68360 microprocessor, 25 MHz, 4MB RAM);

• OS-9 operative system;– Operating as a special purpose

devices:• Implement a standard set of rule of

operations;• pre-defined rules that can be tuned;

It is difficult to operate these systems according to non-standard rules; DTS 170

Traffic control system development process

1. The traffic control engineers develop a new traffic control system design;

2. The new design is given to a vendor to be implemented;

Problems:– What if there is no budget to

contract an EE to develop the system?

– What if the system returned by the vendor does not behave as expected?

Proposed approach: – ENABLE the traffic control

engineers to implement the systems;

– CLOSE the gap between design and implementation;

≠

Proposed approach

• Develop an integrated software suite that automatically translate the design into executable code:– TTCS – Tools for the development of traffic control systems;

• Design / Simulation environment abstract from the low level details:– Control logic can be expressed at high level (as a

mathematical equation for example);– Low level hw related (often platform specific) issues hidden

from the control designer;• The controller logic design is automatically converted

into executable code:– No need to hire electrical engineers to turn the design into

low-level C code;• Potential advantages: Cost reduction, reduction of

level of expertise needed

HW

TTCS – Tools for the development of traffic control system

Design tools

Traffic Simulator

Interface with Simulator

Interface with Control

PC104

2070

Interface with 2070

Executable codeAutomatic compilation

NTCIP

pc104.exe motorola.exe

Design tool: Simulink

• High level language for control design: control laws can be expressed as a difference / differential equation;

• Synchronous abstraction: time is a sequence of instants;

• Interconnected blocks declare the relation between their input and outputs;

/

Source: Yuwei Li, Wei-Bin Zhang, “Summary of requirements: Los Angeles Transit Priority System”, UCBerkeley

Design tool: Simulink

– The developed control system design is then AUTOMATICALLY converted into executable code for target architecture using RealTime Workshop.

– No need to worry about low-level hardware details;– No need to hire an expert for the implementation;

Simulink

Control

System model

Embedded System

Control(Executable code)

RT Workshop

Traffic Simulator / 2070 Interface

In order to design a traffic control system is necessary to test the design in a simulated environment:– Software simulation (e.g. Paramics, Vissim, etc.);– Hardware in the loop simulation (e.g. PATH

arterial traffic/transit lab);

Currently under development:– Interface with the PATH HIL system;– Interface with sw traffic simulator;– Interface with 2070s;

Preliminary results

Simulink 2 Traffic Simulation Interface:– Under development (currently in third beta);– Socket based;– Interface with the traffic simulator used in the PATH traffic lab;

Simulink 2 2070 interface:– Design stage;– NTCIP based;

Simulink(CONTROL)

C(k) = D(k-1) * ….

PATH traffic lab traffic simulator

S2TS

TCP sockets TCP sockets

Traffic Controller implemented so far

Implemented:– pre-timed 2 phases NEMA control;– pre-timed 4 phases NEMA control;– Semi-actuated control;

Currently implementing:– Coordinated pre-timed control;– LADOT ACTS;– Abu-Lebdeh’s Integrated Adaptive-Signal

Dynamic-Speed Control;

Case study 1: pre-timed NEMA ring

cycle_timek = timek mod cycle_length

2 cycle_timeK > cycle_length * phase_split

Phasek = 4 otherwise

MATLABFunction

To Paramics Scope

1

Resolution (sec)

<=

RelationalOperator

Product

0.8

Phase split (main/minor)

4

Phase 4

2

Phase 2

MultiportSwitch

[PHASE_SPLIT]

[RESOLUTION]

[CYCLE_LENGHT]

[PHASE_SPLIT]

[CYCLE_LENGHT]

[RESOLUTION]

60

Cycle length(sec)

Cy cle length

Clock Resolution

Time in cy cle

Cycle Time

Source: “Signalized Intersections: Information Guide. Chapter 4: Traffic Design and Illumination”, USDOT FHWA-HRT-04-091

Case study 2: Semiactuated

Semi-actuated intersection:– Minor road get a green only when actuated– Green on both directions need to last at least min_green seconds;

Phase_ACTk stop_line_sensor_activatedk

Phasek = Phase_NACTk NOT stop_line_sensor_activatedk

2 (Phasek-1 = 2) AND (Main_in_greenk < min_green)

Phase_ACTk = 4 otherwise

MATLABFunction

To Paramics1 Scope1

Green length MINOR

MIN green

Phase(k-1)

Phase NACT

Phase NACT

Green length MAJOR

MIN green

Phase(k-1)

Phase ACT

Phase ACT

MultiportSwitch1

Memory

[STOP_LINE_DETECTOR]

[GREEN_LENGTH_MINOR]

[GREEN_LENGTH_MAJOR]

[MIN_GREEN]Phase(k)

Today’s traffic controllers

Networked devices:– coordinate operations;– remote control:

• National Transportation Communications for ITS protocol (NTCIP) set of standards;

• Universal Traffic Data Format (UTDF);

2070 controller: I/O communication ports and sensors (Ethernet, serial, etc);

It is difficult to operate these distributed systems according to non-standard operation rules;

Distributed system are harder to program: communication, heterogeneity, synchronization, etc issues

Case study: LADOT ATCS approach

• “In the mid-1990s [LADOT] recognized the difficulty to maintain and enhance the system because of the low-level programming language it required”;

• Centralized approach;

Adaptive Traffic Control SystemSource:

Yuwei Li, Wei-Bin Zhang, “Summary of requirements: Los Angeles Transit Priority System”, UCBerkeley

Case study: Southern Queensland (AU)

Scenario:– Small town (20 signals);– 400 Km away from the TMC;– Wired communication

infrastructure too expensive ($1,000/yr*km);

– Wireless (mesh) + DSL (67% savings over initial and operational costs);

Source: ITS International, March/April 2006 issue

Lesson learned

• Wireless is cheap and easy but:– Bandwidth is limited;– Higher data loss;

• Distributed systems are harder to program;• Main gain was obtained adding:

– Flexibility;– Upgradeability;– Examples:

• Switching from dedicated communication networks (e.g. the one used in ACTS) to an open IP architecture;

• Modular code (subsystem can be easily plugged-in / upgraded without impacting the rest of the system);

Source: ITS International, March/April 2006 issue

Research question

Can Simulink system be distributed over communication networks while preserving their semantic (i.e. behave as in simulation) and the modular structure (so that local changes can be handled locally)?

SimulinkControl

Embeeded System (2070)


Embeeded System (PC104)


Control(Executable code)Traffic Simulator



Answer

NO:– No support for code distribution;– Compiled code is NOT modular!

System

2070 controller dependencies[…]

Cycle length control

Phase split control

Offset control

Compilation scheme problem

• Computations (read input, write outputs, internal computations) need to be carried sequentially;

• The compiler fixes a computation order to avoid “deadlocks”;

• Algorithm: choosing any linearization of the I/O causality relation and execute computations in that order;

Input 1

Input 2

Output 1

Output 2





1

43

2

Input 1

Input 2

Output 1

Output 2





1

43

2

Input 1

Input 2

Output 1

Output 2

Formal framework

Step 1: Develop a formal framework where to investigate the question theoretically;

A modified version of the Synchronous Transition System[1] (STS) formalism has been used to model Simulink;

Standard Reactive Automata[2] (RA) formalism has been used to model the executable sequential code;

The semantic is given in terms of traces[3];

[1] Manna, Pnueli, “The temporal logic of reactive and concurrent systems”, Springer-Verlag 1992[2] Caillaud, Caspi, Girault, Jard, “Distributing automata for asynchronous network of processors”,

European Journal on Automated Systems, 1997[3] Benveniste, Caillaud, Guernic, “Compositionality in dataflow synchronous languages:

specification and distributed code generation.” Information and Computation (2000)

First result

We define the map from RA to FSTS traces as Benveniste[3]:

Implementation theorem: given the map from RA to STS* traces the implementation map has the following property: for all STS* s and RA r the following holds:

That is to say the RA r that “implements” the FSTS s has the same set of behaviors of s.

[3] Benveniste, Caillaud, Guernic, “Compositionality in dataflow synchronous languages: specification and distributed code generation.” Information and Computation (2000)

[4] Zennaro, Sengupta, “Distributing Synchronous Systems with Modular Structure”, IEEE CSS Conference on Decision and Control, 2004

[5] Zennaro, Sengupta, "Distributing Synchronous Programs Using Bounded Queues", 5th ACM International Conference on Embedded Software (EMSOFT), 2005

Second result

• Monomorphism: is a monomorphism between (FSTS, xSTS) and (RA, xRA). The following must hold: for all FSTS s1 and s2 and RA r1 and r2 :

[4] Zennaro, Sengupta, “Distributing Synchronous Systems with Modular Structure”, IEEE CSS Conference on Decision and Control, 2004[5] Zennaro, Sengupta, "Distributing Synchronous Programs Using Bounded Queues", 5th ACM International Conference on Embedded Software (EMSOFT), 2005

Second result

That is to say:• Since xRA can be implemented across networks, since

maps FSTS to RA with the same behavior, we can implement an FSTS as a distributed RA system;

• Synchronous program can be implemented across networks taking full advantage of concurrency while preserving the synchronous semantic;

• The implemented FSTS system maintain the modular structure of the original FSTS system; Because of it changes can be handled locally;

From theory to practice

Step 2: Based upon this theoretical framework, we built a library to use with Simulink for the development of Modular Distributed Systems (MDS) library[6];

Source: Zennaro, Sengupta, "Distributing Synchronous Programs Using Bounded Queues", 5th ACM International Conference on Embedded Software (EMSOFT), 2005

Research question

Can Simulink system be distributed over communication networks while preserving their semantic (i.e. behave as in simulation) and the modular structure (so that local changes can be handled locally)? YES

SimulinkControl





Control(Executable code)Traffic Simulator



Registered vehicles in india

0

5000000

10000000

15000000

20000000

25000000

1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995

YEAR

Veh

icle

nu

mb

er

A more sustainable approach to traffic control systems

• Developing countries are most affected by road traffic accidents;

• The available budget is limited;• Experts availability may be limited;

The presented approach: – reduce the development /

deployment / maintenance / upgrade costs;

– reduce the required level of expertise;

Relying on:– Cheaper hw;– Appropriate sw environment;

Computing platforms trend

Technology penetration

Wired and Cell Phones grows in India

0

20

40

60

80

100

120

140

160

1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

Year

Mil

lio

n o

f p

ho

ne

sub

scri

ber

s

Data source: Telecom Regulatory Authority of India

Conclusion

• Modern traffic controllers are sophisticated general purpose computers that are hard to program for non standard operation rules;

• The proposed software environment simplify the development of the system automatically converting the high level design into executable code;

• Modern traffic controllers can be interconnected for remote or coordinate control. Again they are hard to program to follow non standard operation rules;

• Wireless technologies can be used to reduce the system cost;

• Simulink does not support distributed compilation;• The proposed compilation scheme allow distributed

modular compilation of Simulink systems;• The cost and level of expertise reduction makes the

technology more accessible;

Questions?

Software download:• TTCS: http://ttcs.zennaro.net

Related publications:• Zennaro, Sengupta, “Modular Composition of Synchronous

Programs: Applications to Traffic Signal Control”, submitted to ACM Transaction on Embedded Computing Systems.

• Zennaro, Sengupta, “Distributing Synchronous Programs Using Bounded Queues, a coordinated traffic signal application”, University of California at Berkeley, Intelligent Transportation Studies, UCB-ITS-RR-2005-4, May 2005

• Zennaro, Sengupta, “Distributing Synchronous Programs Using Bounded Queues” , 5th ACM International Conference on Embedded Software (EMSOFT'05), December 2005

• Zennaro, Sengupta, “Distributing Synchronous Systems with Modular Structure”, IEEE 2004 44th Conference on Decision and Control, December 2004

http://ttcs.zennaro.net/

Extra slides

Just in case…

“Western” solutions

Increasing system complexity:• single traffic light operating

according to pre-timed plans;• actuated system able to sense

and respond to traffic conditions;

• coordinated intercommunicating traffic lights along an arterial;

• systems able to accommodate different traffic priorities.

Previous experiences: lesson learned

Previous experiences:– Computers;– Computer networks;– Telephones;

Similar trends:– Started as:

• expensive solutions• shared by elite experts

– To became:• affordable mass gadgets• Used everywhere around the world (120 mil cells in

India…)

What made this possible?

time

Ope

rato

rs /

Dev

ice

1946

ENIAC

ENIAC (1946)

Cost: $500,000

Intel 4004

Mainframe

(costly, tens of operators per machine)

Minicomputer

(less costly, ten operators per machine)

Microcomputer

(affordable, few operators per

machine)

1970

Personal computer

(affordable, one operator per

machine)

today

PDAs, …

(many machines per person)

Apple II

IBM 5150

Lotus 123

Apple NewtonIPOD

time

Ope

rato

rs /

Dev

ice

1946

ENIAC

Intel 4040 (1971)

Cost: $1000

Intel 4004

Mainframe


Minicomputer


Microcomputer


machine)

1970

Personal computer


machine)

today

PDAs, …


Apple II

Apple 2 (1977)

Cost: $1298

IBM 5150

Lotus 123

Apple NewtonIPOD

time

Ope

rato

rs /

Dev

ice

1946

ENIAC

IMB 5150 (1981)

Cost: $1,565

Intel 4004

Mainframe


Minicomputer


Microcomputer


machine)

1970

Personal computer


machine)

today

PDAs, …


IBM 5150

Apple II

Lotus 123

Apple NewtonIPOD

time

Ope

rato

rs /

Dev

ice

1946

ENIAC

Intel 4004

Mainframe


Minicomputer


Microcomputer


machine)

1970

Personal computer


machine)

today

PDAs, …


Apple II

Apple NewtonIPOD

IBM 5150

Lotus 123

Apple Newton (1993)

Cost: $1000

Apple IPOD (2001)

Cost: $299

Outline

• Problem statement;

• Lesson learned from similar experiences;

• Case study: Southern Queensland Traffic Control System;

• Proposed approach;

• Achievements;

• Current and future work;

What made this possible?

System life-cycle:– Cost reduction in the initial investment:

• ENIAC: $500,000;• PC: <$1,000;

– Simplified deployment:• ENIAC: team of scientist;• PC: single operator (undergrad);

– Simplified maintenance: • ENIAC: team of scientist;• PC: automatic / single admin can monitor multiple

systems;– Simplified upgrade:

• ENIAC: system shutdown;• PC: plug-and-play;

Research problem

This is due to the existing gap between:– Simulation environment:

• Used during the design phase;• Must hide implementation details, traffic control designer

should not be concerned of low level hardware details;

– Implementation environment:• Used in the development phase;• Need to address low level details;

GOAL: close the gap between design and implementation

Road Traffic Injury Problem

Problem size:– 1.2 million death worldwide;– 20 million injured people

Disparity between high-income country and the rest of the world:– Less than 130,000 fatalities in high income

country;– Second cause of death on the 5 to 29 y.o.

population in low and middle income country;– Car accident ranks 8th in the global burden of

disease and injuries (DALYs scale);

Source: Pedan et all, “The injury Chart Book”, WHO, Geneva 2002

Transportation Trend

Data source: http://www.photius.com

Registered vehicles in india

0

5000000

10000000

15000000

20000000

25000000

1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995

YEAR

Veh

icle

nu

mb

er

Accidents trend

Source: A. Sarna, “Improving Road Safety in Developing Countries Workshop” presentation, 2006

Road Traffic Injury Problem

Source: Mathers et al. “Updated projections of Global mortality and Burden of Disease”, WHO 2005

Traffic control system design process

1. The traffic control engineers develop a new traffic control system design







Problems:– What if there is no budget to

contract an EE to develop the system?

– What if the system returned by the vendor does not behave as expected?

Proposed approach: – ENABLE the traffic control

engineers to implement the systems;

– CLOSE the gap between design and implementation;

≠

The vision

Simulink

Control

Simulatedsystem







RT Workshop +Our libraries

Use the same development process to design the control code and compile into code to be executed on the traffic controllers.

Achievements

• We implement synchronous subsystems as asynchronous reactive automata with an equivalent behavior;

• We compose automata using rendezvous composition and we prove that every behavior of the resulting system can be mapped to a behavior of the original system;

vk := uk wk := vk

uk vk wk

?u(su)

sv:=su;

!v(sv)

u0=0

u1=1

u3=2

w0=0

w1=1

w3=2

v0=0

v1=1

v3=2 ?v(sv)

sw:=sv;

!w(sw)

Achievements

vk := 2 * uk wk := vk+zk

ukvk wk

• The proposed approach take full advantage of concurrency while hiding the complexity of the interleaving from the user;

• We proved that the proposed approach maps causal-loop free synchronous systems into dead-lock free asynchronous systems;

zk := 2 * uk

zk

u0=0 … w0=0

u1=1 … w1=4

u2=2 … w2=8

?u(su)

sv:=2su;

!v(sv)

?u(su)

sz:=2su;

!z(sz)

?v(sv)?z(sz)

?z(sz) ?v(sv)

sw:=sz+sv;!w(sw)

Achievements


ukvk wk

• A local change requires only local re-compilation (i.e. the modularity of the synchronous system is preserved in its asynchronous equivalent);

zk := 2 * uk

zk

u0=0 … w0=0

u1=1 … w1=4

u2=2 … w2=8

?u(su)

sv:=2su;

!v(sv)

?u(su)

sz:=2su;

!z(sz)

?v(sv)?z(sz)

?z(sz) ?v(sv)

sw:=sz+sv;!w(sw)

Achievements


ukvk wk

• Local change requires only local re-compilation (i.e. the modularity of the synchronous system is preserved in its asynchronous equivalent);

zk := 3 * uk

zk

u0=0 … w0=0

u1=1 … w1=4

u2=2 … w2=8

?u(su)

sv:=2su;

!v(sv)

?u(su)

sz:=3su;

!z(sz)

?v(sv)?z(sz)

?z(sz) ?v(sv)

sw:=sz+sv;!w(sw)

Case study: ASDS control system

Step 3: Implement some simple traffic control system as a proof of concepts

Adaptive-Signal Dynamic Speed Control:– GOAL: minimize delay and number of stops;– IDEA:

• treat speed as a control variable;• drivers’ choice is link optimal while ASDS can select a system-optimal

speed;

Source: Abu-Lebdeh, “Integrated Adaptive-Signal Dynamic-Speed Control of Signalized Arterials", 5th ACM International Conference on Embedded Software (EMSOFT), 2005

Case study: ASDS control system

The alternative control scheme was developed using Simulink and run over wirelessly connected laptops;

Case study 1: Paramics

Case study: Simulink

MATLABFunction

To Paramics

Cy cle length

Clock Resolution

Time in cy cle

Time within Cycle

Scope

1

Resolution (sec)

0.7

Phase split (main/minor)

Cy cle length

Phase split

Time in cy cle

Phase number

Phase selection

3

Phase 3

1

Phase 1

MultiportSwitch

[PHASE_SPLIT]

[RESOLUTION]

[CYCLE_LENGHT]

[PHASE_SPLIT]

[CYCLE_LENGHT]

[RESOLUTION]

[CYCLE_LENGHT]

60

Cycle length(sec)

1

Phase numberSign

max

MinMax

f(u)

Distance from phase change

0

Constant

3

Time in cycle

2

Phase split

1

Cycle length



Case study 3: Simulink

MATLABFunction

To Paramics

Scope

1

Resolution (sec)

Det 1

Det 2

Green length M

MinGreen

Green length m

Phase selected

Phase selection logic

2

Phase 3

1

Phase 1

MultiportSwitch

15

Min green (sec)

Phase

Clock Resolution

Current green length MAJOR

Current green length minor

Green length



[GPS_TIME]

[SL_DETECT_2]

[SL_DETECT_1]

[PHASE_TO_PAR]

[MIN_GREEN]

[RESOLUTION]

[CYCLE_LENGHT]


[MIN_GREEN]

[SL_DETECT_2]

[SL_DETECT_1]

[PHASE_TO_PAR]


[PHASE_TO_PAR]

[RESOLUTION]

60

Cycle length(sec)

Major Phase

Minor phase if actuated

Signal Phase

GPS_time

StopLine Detector #1


MATLABFunction

To Paramics

Scope

1

Resolution (sec)

Det 1

Det 2

Green length M

MinGreen

Green length m

Phase selected

Phase selection logic

2

Phase 3

1

Phase 1

MultiportSwitch

15

Min green (sec)

Phase

Clock Resolution

Current green length MAJOR

Current green length minor

Green length



[GPS_TIME]

[SL_DETECT_2]

[SL_DETECT_1]

[PHASE_TO_PAR]

[MIN_GREEN]

[RESOLUTION]

[CYCLE_LENGHT]


[MIN_GREEN]

[SL_DETECT_2]

[SL_DETECT_1]

[PHASE_TO_PAR]


[PHASE_TO_PAR]

[RESOLUTION]

60

Cycle length(sec)

Major Phase

Minor phase if actuated

Signal Phase

GPS_time



Case study 3: Simulink

1

Phase selected

min

min

z

1

Unit Delay2

z

1

Unit Delay1

Sign4

Sign3

Sign2

Sign1

-1

Phase 6

1

Phase 5

0

Phase 4

0

Phase 3

0

Phase 2

0

Phase 1

max

Min2

min

Min1

max

Max1

max

Max

Add5

Add4

Add3

Add2

Add1

Add

5

Green length m

4

MinGreen

3

Green length M

2

Det 2

1

Det 11 if there is v ehicle

1 if min green not satisf ied

time

Ope

rato

rs /

Dev

ice

1946

ENIAC

ENIAC (1946)

Cost: $500,000

Intel 4004

Mainframe


Minicomputer


Microcomputer


machine)

1970

Personal computer


machine)

today

PDAs, …


Apple II

IBM 5150

Lotus 123

Apple NewtonIPOD

Apple IPOD (2001)

Cost: $299

Technology penetration

Wired and Cell Phones grows in India

0

20

40

60

80

100

120

140

160

1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

Year

Mil

lio

n o

f p

ho

ne

sub

scri

ber

s

Data source: Telecom Regulatory Authority of India

Toward a sustainable approach to traffic control systems

• Sustainable approach:– Abating the cost in each phase of the system

life cycle:• Leveraging on embedded computing and wireless

technologies (cost reduction)

– Abating the level of expertise required to develop / deploy / maintain the system (simplification):

• Leveraging on suitable software environments;

HW

TTCS – Tools for the development of traffic control system

Design tools

Traffic Simulator



PC104

2070

Interface with 2070

Executable codeAutomatic compilation

NTCIP

1.exe 2.exe

Case study 2: Utah intersection

• Multi lane intersection (somewhere, Utah);• Pre-timed 4-phases NEMA ring

Documents

Toward a new generation of traffic control systems Marco Zennaro Prof. Raja Sengupta