Prof. dr. ir. Serge P. Hoogendoorn Technische Universiteit Delft, AMS, Arane

Future of Traffic Management and ITSPutting the ‘I’ in ITS

Societal urgency Regional Traffic Management • Urbanisation is a global trend leading to

higher concentration of people and their movements

• Accessibility is a major issue in many car-centric cities and appears to worsen further in coming decades

• 2/3rd of traffic accidents occurs within city boundaries

• High impact (traffic-related) emissions and noise (people live near roads…)

• Urban space is very scarce, so building new infrastructure is generally not easy / cheap

• Focus of today’s lecture is on improving utilisation of existing infrastructure by smart management of traffic flows

26%38%

33%

Why does it make sense to manage traffic flows?

• Wide Moving Jam (often also called: start-stop wave - in Dutch: ‘filegolf’)

• Occurs ‘spontaneously’ in unstable flow; its occurrence is hence very hard to predict

• Once present, a wide moving jam… - Reduces road capacity substantially

- Has very predictable dynamics (moves at 18 km/h in opposite direct of traffic)

- Increases un-safety, pollution, and fuel consumption

• In some cases, the trigger is more clear…

Traffic stability & Moving jams

Jams @ sags• 50% congestion in Japan originate at sags

• Empirical analysis on behalf of Toyota showed that changes in car-following behaviour is the main cause

• Instability of flow causes start-stop waves to be emitted from sag area

• Substantial reduction of capacity results

• Resolving them yield substantial improvement in throughput!

To make matters worse…• Recent empirical work shows relation between

speed in jam and jam outflow

• Relation shows that outflow of wide moving jams in 30% lower than (free) road capacity

• Congestion leads to ‘more congestion’… - Capacity reduction leads to higher delays

- WFMs trigger new standing queues - WMFs occur start in standing queues

• But also congestion spill-back strongly reduces throughput

Start-stop wave reducescapacity by 30%

Standing queue reduces capacity by 15-25%

• Next to reduced capacity in case of standing queues or wide-moving jams, severe reduction of throughput may be cased by spill back

• Example shows how spill back from on-ramp reduces outflow of network substantially…

• Spillback occurs at off-ramps, urban arterials, etc.

Spillback causes 245 minutes of avoidable travel

delays!

The Hype: NFD’s

Yokohama

San Francisco

Nairobi

0 50 100 150 200 250Density

0

1000

2000

3000

4000

5000

6000

7000

8000

Flow

MFD data v2Amsterdam?

• NFD: Network Fundamental Diagram

• Describes average throughput of network als a function of network load (e.g. average density)

• Provides insights into functioning of networks and a way to cross-compare networks (see illustration)

• Important characteristics: tipping point (critical density) from which point onward the production reduces or even becomes zero

• Why the big differences between networks?

Network level impacts

AV E R A G E D E N S I T Y I N N E T W O R K

EXIT

RAT

ES

• Generalised NFD shows impact of uneven distribution of traffic over network

• Empirical research shows how a more even distribution of density over network leads to substantial increase in network performance

• Improvements up to 30% in throughput seem possible!

• g-NFD can be derived from underlying FDs (example shows Greenshields)

• Also holds for other networks, including pedestrians! Q

network

(⇢,�) = Qlocal

(⇢)� v0

⇢j

�2

10

Prevent blockades, e.g. by increasing queue outflow, or separating flows in different directions / use of reservoirs

Distribute traffic over available infrastructure for instance by means of guidance or intersection controllers (BP)

Increase throughput in particular at pinch points in the network…

Limit the inflow (gating) ensuring that number of vehicles / pedestrians stays below critical value (NFD)!

Traffic Management by First Principles • Insights into the causes of

operations deterioration have led to the development of simple but useful principles of traffic control and management

• Principles have formed basis for designing traffic control and management plans, altering network desing, etc.

Applications are not restricted to car traffic…

General idea: temporarily limit the inflow into the WMJ using reduced speed limits upstream of jam…

1. Detect the wide moving jam using inductive loops 2. Determine its severity (number of ‘excess’ vehicles to be

‘removed’ by limiting inflow)

3. Determine if there is space to temporarily store vehicles that are witheld to flow into jam

4. If solvable (available space > severity), implement control strategy

5. Monitor to check if jam is resolved

Back to the wide-moving jam…

x x x

t

Some traffic engineering background• How to effectively limit inflow using speeds: difference between local

deployment and instantaneous deployment

• Which of the schemes is (most) effective?

Tplatoon

TplatoonT

platoon

no limit local limit instant. limit

t t

Isolated approach using VSL• SPECIALIST algorithm developed by TU Delft on behalf of RWS

• Fixed speed limit deployed over variable roadway stretch: SPECIALIST computes length that is required to remove excess vehicles

• After tuning, we had 2.8 activations per day resolving jam in 72% cases…

Succes? Saves approx.

700k AUD annually

Improvements?Why not more activations? Why no “only” 72% solved?

Improving effectiveness?1. Increase # activations by increasing control

space: support variable speed limits by using other control measures, e.g. ramp metering: coordination

2. Increase % of WMJ resolved upon activation by correctly determine control task (vehicles ‘too many’ in moving jam) and available control space: improved state estimation

3. Only deploy in case moving jam: improving diagnostics

Increasing control space• COSCAL v2 integrates VSL and ramp metering

• Use of ramp as buffer to support VSL control approach making it much more effective

• Shows need for coordination of measures

• Extension to multiple on-ramps, intersection controllers, etc.

• Requires insight into storage space on ramps / intersections and relation with bottleneck

Moving jam Relative space of Slave buffer = relative space of Master buffer

Car as sensor• Floating Car Data to detect WMF (for queue

tail warning application, AID)

• Use of Flitsmeister app to collect data (increased polling in pilot area)

• Comparison shows that in many instances, FCD is more timely and accurate than loop data

• Allows improved differentiation between types of congestion

• Works at moderate penetration levels (4-10%)

• Courtesy of RWS (Marco Schreuder)Loop-based signal

between loops FCD is more timely

Queue estimation• Estimation of queue lengths (nr. vehicles in

queue) from loops isimportant for traffic control and management purposes, but hard problem in traffic engineering

• Determining queue lengths is difficult due to measurement errors inflow and outflow (drift)

• Use of TomTom FCD data (sparse) providing segment travel times

• Machine learning using structure based on problem characteristics

• Machine learning approach outperformed traditional model-based approaches

• Example shows queue estimates for s106 urban arterial

• Quality sufficient for traffic management • Also without induction loops, queue estimates

approach very reasonable

06:00 06:30 07:00 07:30 08:00 08:30 09:00 09:30 10:00tijd Jul 11, 2016

0

100

200

300

400

wac

htrij

(m)

waargenomengeschat

15:00 15:30 16:00 16:30 17:00 17:30 18:00 18:30 19:00

tijd Jul 11, 2016

0

100

200

300

400

wachtrij (m)

waargenomengeschat

06:00 06:30 07:00 07:30 08:00 08:30 09:00 09:30 10:00tijd Jul 19, 2016

0

100

200

300

400

wac

htrij

(m)

waargenomengeschat

15:00 15:30 16:00 16:30 17:00 17:30 18:00 18:30 19:00tijd Jul 19, 2016

0

100

200

300

400

wac

htrij

(m)

waargenomengeschat

Pilot Amsterdam• First practical pilot with coordination

applied on a network level

• The golden triangle pulled it off: road authority, industry and academia

• Pilot showed that coordination - if done correctly - greatly increases effectiveness of traffic management

• Assessment shows positive impacts (100-250 veh-h savings per peak) equal to up to 1 million AUD / year

• In era of Google and Teslas, when do we stop investing in road-side technology?

A simple quiz• Case with 2 OD pairs and perfectly informed

traveler from A to B having 2 options • Intelligent intersection controller optimises

locally capacity use at intersection evenly (equal delays for both directions)

• Most travellers (85%) choose route 2 under normal circumstances

• Event (incident) occurs; flow conditions route 2 worsen (from 120 to 80 km/h)

• What happens?

A

C

D

BRoute 1

Route 2

perc. choice route 1

traffi

c co

ntro

ller

total delays

0 0.1 0.2 0.3 0.40

0.1

0.2

0.3

0.4

0.5

1

2

3

4

5

6

x 105

Cooperation road-side & in-car• Consideration of drivers changing route due to

changes in traffic operations is necessary

• Joint control of intersection and route guidance leads to System Optimum (SO), but may not be feasible

• In designing traffic control, take into consideration expected impact the controller actions have on routing (anticipatory control)

• Requires sending information about drivers route choice to traffic controller (“V2I”)

• First practical tests (PPA) show technical feasibility, but not yet network impacts

1

No consideration of route demand changes

in control

Anticipating demand changes

Towards car-based traffic management

The car as an actuator… • Can we use the car as an actuator to improve traffic conditions?

• Naive approach: an automated car drivers (potentially?) at a shorter time headway (either in a platoon or not) and hence the road capacity increases

• Relatively small increases in capacity (e.g. 2.6% in case of 10% penetration of vehicles driving witg 25% smaller headway)

• Will this be feasible in mixed traffic?

• Can we do better? Work on behalf of Shell has focussed on the improvement of traffic flow (for either throughput or emissions) when penetration levels are still well beyond 50%….

MPC for cooperative driving • Consider state of platoon

• We control acceleration of some veh.

• Predict behaviour non-controlled vehicle

• Idea is to find control functional that minimises some objective function (delays, fuel, emissions, comfort) describing costs (performance) for entire platoon

• Recompute optimal control function for new time instant

Follower 2 -Human-driven

vehicle

Follower 1-Cooperative vehicle

Leader – Human-driven

vehicle

s1, Δv1 s2, Δv2 ~x(t) = {ri(t), vi(t)}

i 2 Cui

uj = a

1�

⇣ vjv0

⌘��✓s⇤(vj ,�vj)

sj

◆2!

j 2 U

~u[tk,tk+H)

~u⇤[tk,tk+H) = argmin J(~u[tk, tk + T ))

tk+1

MPC for cooperative driving • Impact of C-ACC on

wide-moving jams

• Lower penetration (e.g. 10%) jams are resolved effectively

• Possible impact of cooperative systems in transition to full penetration

Use of game theory for C-ACC systems• Differences no control, ACC and C-ACC

• Cost function in C-ACC case describes cost of entire platoon, controlled vehicle influences behaviour of the vehicles to minimise overal costs

More info: Wang, M., Hoogendoorn, S.P., … (2015). Game theoretic approach for predictive lane-changing and car-following control (2015) TR-Part C: Emerging Technologies, 58, pp. 73-92.

Any downsides to this approach?

Focus on decentralisation of control approaches to

deal with complexity!

Jams @ sags: resolution• Similar approach to reduce congestion

caused by sags

• Controlled vehicle does not optimise local conditions, but tries to stabilise flow so congestion + cap drop is smaller

• Non-trivial but understandable primary (DADA) and supporting strategies result…

• Substantial improvements by controlling only a few vehicles

More info: Goñi-Ros B. et. al (2016) Optimization of traffic flow at freeway sags by controlling the acceleration of vehicles equipped with in-car systems, TR-C: Emerging Technologies, 71, 1-18

Jams @ sags: resolution• Substantial improvements can be achieved by controlling only a few vehicles!

NIVEAU 0

DELEN IN BLOEI

VAN NIET DELEN NAAR DELEN

EVOLUTIE VAN DE PRIVÉAUTO

NIVEAU 1-2

NIVEAU 3-4

NIVEAU 5

Mens en machine Coöperatief rijden Gemengd verkeer Stedelijk dilemma Zelfrijdende stad

Mens en machine Coöperatief rijden Gemengd verkeer Stedelijk dilemma Zelfrijdende stad Aanpassen Verdrag van Wenen Toelating voertuigen niveau 1-2 Aansprakelijkheid en

verzekerbaarheid Vereisten rijbewijs Human-machine interface

Data & privacy Toelating coöperatieve voertuigen Internationale coördinatie

coöperatief rijden

Ethische issues zelfrijdende voertuigen

Toelating voertuigen van niveau 3-4

Internationale coördinatie niveau 3-4

Minimale volgafstanden Regelgeving in-/uitvoegen

Regelgeving en handhaving voor veilig en vlot rijden niveau 3-4 in stad

Toelating voertuigen niveau 5 Regelgeving voor veilig en vlot

rijden niveau 5 in stad Internationale coördinatie (niveau 5) Regelgeving eerlijke concurrentie

(bij delen)

Wegbelijning en bebording op orde Adaptieve planning en contracten

(innovaties) Pilots coöperatief en niveau 3-4

Investeren in V2I, V2C Opschalen pilots niveau 3-4 op

snelweg Testen veiligheid in gemengd

verkeer

Wel/niet niveau 3-4 op aparte rijstroken

Wel/niet breedte rijstrook aanpassen

Pilots niveau 5 stad

Routes in steden voor niveau 3-4 aanpassen

Wegbelijning en bebording op orde in stad

Opschalen niveau 5 stad

Concrete maatregelen voor niveau 5 in stad

Benodigde parkeerruimte (als delen een vlucht neemt)

Human-machine interface Criteria rijbewijs Veiligheid coöperatieve systemen Attitudes consument coöperatief

rijden Benodigde digitale infrastructuur

(V2I, V2C)

Ethische issues zelfrijdende voertuigen

Veilige volgafstanden gemengd verkeer

Benodigde breedte rijstroken Niveau 3-4: comfort en

wagenziekte

Wens consument om binnen stad automatisch te rijden

Benodigde ingrepen en kosten niveau 4 in stad

Wanneer niveau 5?

Maatregelen niveau 5 in stad: Fysieke scheiding vervoerwijzen

Slimme camera’s en sensoren Lage snelheid in stad

‘Dwingende’ voertuigen

Ontwikkeling delen: naar ‘Delen in bloei’?

Verkoop en penetratie niveau 1-2-systemen

Houding en acceptatie burger (niveau 1-2)

Maatschappelijke effecten niveau 1-2 buiten stad

Effecten pilots (niveau 3-4)

Verkoop en penetratie niveau 1-2 (coöperatieve) systemen

Houding en acceptatie burger (coöperatief rijden)

Snelheid ontwikkeling niveau 3-4-technologie

Effecten pilots opschaling (niveau 3-4)

Verkoop en penetratie niveau 3-4-systemen

Houding en acceptatie burger (niveau 3-4)

Maatschappelijke effecten niveau 3-4 in de praktijk

Snelheid ontwikkeling niveau 5-technologie

Effecten pilots (niveau 5)

Houding en acceptatie burger (niveau 3-4 stad)

Maatschappelijke effecten niveau 3-4 in de stad

Snelheid ontwikkeling niveau 5-technologie

Effecten pilots en opschaling (niveau 5)

Ontwikkeling delen: naar ‘Delen in bloei’?

Verkoop en penetratie niveau 5-systemen

Houding en acceptatie burger (niveau 5)

Maatschappelijke effecten niveau 5 (vooral in stad)

16#487 KiM_paden naar een zelfrijdende toekomst_omslag_def.indd 6-10 06-03-17 14:47

Cooperative driving Mixed traffic Urban dilemma Zelf-driving cityMan-machine

(c) KiM, The Netherlands

Will man and machine work well

together?

Will cooperative systems turn out to be feasible (e.g. due

to privacy, security)

Will we be able to deal with mixed

traffic?

Will self-driving vehicles need

separate infrastructure?

Will self-driving vehicles interact well

with vulnerable road users

What about the other modes?

Issues in dense cities are not limited to cars…

Bike congestion causing delays and hindrance

Overcrowding during events and regular situations also due to tourists

Overcrowded public transport hubs

Not-so-seamless public transport

Bike parking problems & orphan bikes

Bike congestion causing delays and dangerous behaviour at intersections

Example trajectories + local densities on Utrecht platform

Example of possible intervention showing potential impact of station crowd management

Managing Station Pedestrian Flows • Dutch railway (ProRail and NS) with

support of TU Delft have been working on SmartStation concept

• Multi-level data collection system • Detailed density collection at pinch

points (e.g. platforms) • WiFi / BlueTooth at station level • Combination with Chipcard data

provides comprehensive monitoring information for ex-post assessment and real-time interventions

Bike Traffic Management? • Different examples of bike

traffic management, such as bike parking information Utrecht and dynamic routing are piloted

• Joint work of TU Delft and TNO showed potential of combining speed advice (e.g. via app, or via lights) and green waves (reduction of #stops of 45%)

• Potential for effective approaches increases with increased connectivity

The Dutch alternative to the self-driving car?

Change in research focus… Towards Smart Urban Personal Mobility

37Regional traffic

management & controlFlexible public

transport servicesUrban active mode

mobilityCooperative systems

and driver automationUrban Traffic and

Transport data

Thank you for your attention

Traffic management in transition• Traffic Management has been successful in better utilising existing

infrastructure, recent major advances in coordination of measures • New in-car technology has potential to make traffic management more effective

by improving monitoring and actuation, even at low penetration levels, if (and only if!) we use smart approaches to make most out of the new options!

• Important aspects in this are the speed at which cooperative driving will be introduced and how we deal with mixed traffic

• But there are other aspects that will determine the speed at which we move towards fully automated urban transportation systems…

Integrated & hyper-connected urban mobility

• Uni-modal urban transport system not likely!• Using key technological trends (big data,

connectivity), social trends in attitude towards (car-) ownership, increased flexibility in work and leisure time, and objectives / requirements regarding urban mobility (impacts of liveability, health, and resilience)…

• Innovations should foster transition to a integrated connected urban mobility system, with pillars:

1. Seamless integration of services - prioritising sustainable modes - via hyper-connectivity

2. Flexible / efficient use infrastructure & space3. Common Data Platform

40