A Distributed and Self-Regulating Approach for Organizing a Large - - PowerPoint PPT Presentation

A Distributed and Self-Regulating Approach for Organizing a Large System of Mobile Objects

S´ andor P. Fekete Stefan Fischer Bj¨

• rn Hendriks

Horst Hellbr¨ uck

Algorithms Group, Department of CS

Sebastian Ebers

Institute of Telematics Universit¨ at zu L¨ ubeck

October, 8th 2010

Review Hovering Data Clouds Improving Flow Urban traffic Outlook

Goal: Improving traffic

Cars as self-organizing entities Ad-hoc C2C communication Distributed algorithms Avoidance of centralized coordination and organization

2 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook

Rules, Data and Strategies

Hovering Data Cloud (HDC): Mix of system

• f rules and data

structure which is independent of physical carriers (e.g., vehicles) Organic Information Complex (OIC): Aggregation of information of several HDCs Advanced Distributed Strategy (ADS): Strategies pursued by equipped vehicles to deal with detected traffic phenomena

3 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook

AutoNomos architecture

HDC Layer OIC Layer

Aggregation into higher-level-data Local Sensor Data e.g. GPS, speed, acceleration, ABS, ESP

Car-to-Car Communication

Data mining Detection of matching HDCs Adaptation

• f Driving

Behavior Maintaining HDC Repository

AutoCast Transport Layer

Data Units Information Extraction Correlation Handling 4 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Implementation of HDC and OIC layers

Implementation of HDC and OIC layers

HDC Layer OIC Layer

Aggregation into higher-level-data Local Sensor Data e.g. GPS, speed, acceleration, ABS, ESP

Car-to-Car Communication

Data mining Detection of matching HDCs Adaptation

• f Driving

Behavior Maintaining HDC Repository

AutoCast Transport Layer

Data Units Information Extraction Correlation Handling 5 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Implementation of HDC and OIC layers

Class diagram

+registerListener(ein listener : HDCDataUnitListener) +onReceive(ein dataUnit : HDCDataUnit) +distribute(ein dataUnit : HDCDataUnit) TransportLayer +update(ein event : UpdateEvent) «interface» HDCUpdateListener +update(ein event : UpdateEvent) «interface» SensorUpdateListener +getValue() : object +registerListener(ein listener : SensorUpdateListener) Sensor +correlate(ein data : HDCDataUnit) : HDC #integrate(ein data : HDCDataUnit, ein hdc : HDC) #aggregate(ein hdcs... : HDC) #informationExtraction(ein hdc : HDC) HDCManager +hashValue() : long HDCDataUnit

• ttl : long
• maxLifetime : long
• creationTime : long
• lastAccess : long
• expansion : object
• name : string
• distributionArea : object

+doUpdate(ein dataUnit : HDCDataUnit) +registerListener(ein listener : HDCUpdateListener) «interface» HDC +add(ein dataUnit : HDCDataUnit) «interface» HDCDataUnitListener 1 1 1 * 1 1 1 * * * 1 * * *

6 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Implementation of HDC and OIC layers

Receiving a message

TransportLayer

• nReceive(HDCDataUnit)

HDCManager add(HDCDataUnit) correlate(HDCDataUnit) integrate(HDCDataUnit, HDC) HDC update(UpdateEvent) informationExtraction(HDC) distribute(HDCDataUnit) doUpdate(HDCDataUnit) HDCDataUnit 7 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Implementation of HDC and OIC layers

Receiving a message

TransportLayer

• nReceive(HDCDataUnit)

HDCManager add(HDCDataUnit) correlate(HDCDataUnit) integrate(HDCDataUnit, HDC) HDC update(UpdateEvent) informationExtraction(HDC) distribute(HDCDataUnit) doUpdate(HDCDataUnit) HDCDataUnit

<< HDCDataUnitListener >>

7 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Implementation of HDC and OIC layers

Receiving a message

TransportLayer

• nReceive(HDCDataUnit)

HDCManager add(HDCDataUnit) correlate(HDCDataUnit) integrate(HDCDataUnit, HDC) HDC update(UpdateEvent) informationExtraction(HDC) distribute(HDCDataUnit) doUpdate(HDCDataUnit) HDCDataUnit

<< HDCUpdateListener >>

7 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Avoiding traffic jam with Jam-ADS Improve flow through successive traffic lights

Review of Jam-ADS

Goal Increase traffic flow on a highway while decreasing the fuel consumption Idea Recommend a velocity which is a convex combination of desired velocity and average velocity of the cars ahead vrecommend = λvdesired + (1 − λ)vaverage

8 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Avoiding traffic jam with Jam-ADS Improve flow through successive traffic lights

Patent granted

Our patent application for the Jam-ADS procedure was successful. Granted on 2010-09-09

9 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Avoiding traffic jam with Jam-ADS Improve flow through successive traffic lights

Scanning parameter space

We still run many simulations with variations of independent parameters car density, number of lanes λ, distance to take average

• ver, consider neighbor lanes

penetration (fraction of equipped cars)

10 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Avoiding traffic jam with Jam-ADS Improve flow through successive traffic lights

Scanning parameter space II

Applying different simulators

• ur own simulator

SUMO together with the network simulator Shawn to find capabilities and limits of Jam-ADS.

11 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Avoiding traffic jam with Jam-ADS Improve flow through successive traffic lights

Scanning parameter space III

Average fuel consumptions

12 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Avoiding traffic jam with Jam-ADS Improve flow through successive traffic lights

Improve flow through successive traffic lights

Advanced Distributed Strategy to improve traffic flow over successive traffic lights

13 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook Avoiding traffic jam with Jam-ADS Improve flow through successive traffic lights

General idea

Traffic Lights-Advanced Distributed Strategy Cars receive information

from traffic lights: next cycles from other cars: positions and velocities

Cars estimate arrival time at stop line considering other cars ⇒ Adapt velocity ⇒ Save time and fuel by crossing the stop line with maximal speed

14 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Going beyond

Up to now we have improved the flow within a road. Now we look at the road network as a whole.

15 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Max-flow problem

Offline optimization on a road network Given roads (edges) with known capacities ue Given a fixed flow demand from a source s to a sink t We know algorithms to optimally distribute flow, e.g., augmenting paths

s v1 v2 v3 t

(1, 3) (1, 2) (2, 1) (2, 1) (1, 3) (1, 1)

16 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Dynamic flows

Moreover we know algorithms for flows with limited transit times τe on each edge

s v1 v2 v3 t θ = 0

(1, 3) (1, 2) (2, 1) (2, 1) (1, 3) (1, 1) x = 0 x = 0 x = 0 x = 0 x = 0 x = 0 x = 0 x = 0

Demand 6 flow units from s to t Goal minimize time Result need 5 time units Furthermore the transit times τe could be load-dependent ⇒ NP-hardness

17 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Dynamic flows

Moreover we know algorithms for flows with limited transit times τe on each edge

s v1 v2 v3 t θ = 1

(1, 3) (1, 2) (2, 1) (2, 1) (1, 3) (1, 1) x = 2 x = 0 x = 0 x = 0 x = 0 x = 0 x = 1 x = 0

Demand 6 flow units from s to t Goal minimize time Result need 5 time units Furthermore the transit times τe could be load-dependent ⇒ NP-hardness

17 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Dynamic flows

Moreover we know algorithms for flows with limited transit times τe on each edge

s v1 v2 v3 t θ = 2

(1, 3) (1, 2) (2, 1) (2, 1) (1, 3) (1, 1) x = 1 x = 1 x = 0 x = 0 x = 1 x = 0 x = 1 x = 1

Demand 6 flow units from s to t Goal minimize time Result need 5 time units Furthermore the transit times τe could be load-dependent ⇒ NP-hardness

17 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Dynamic flows

Moreover we know algorithms for flows with limited transit times τe on each edge

s v1 v2 v3 t θ = 3

(1, 3) (1, 2) (2, 1) (2, 1) (1, 3) (1, 1) x = 0 x = 0 x = 1 x = 0 x = 1 x = 1 x = 1 x = 1

Demand 6 flow units from s to t Goal minimize time Result need 5 time units Furthermore the transit times τe could be load-dependent ⇒ NP-hardness

17 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Dynamic flows

Moreover we know algorithms for flows with limited transit times τe on each edge

s v1 v2 v3 t θ = 4

(1, 3) (1, 2) (2, 1) (2, 1) (1, 3) (1, 1) x = 0 x = 0 x = 0 x = 1 x = 0 x = 1 x = 0 x = 1

Demand 6 flow units from s to t Goal minimize time Result need 5 time units Furthermore the transit times τe could be load-dependent ⇒ NP-hardness

17 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Dynamic flows

Moreover we know algorithms for flows with limited transit times τe on each edge

s v1 v2 v3 t θ = 5

(1, 3) (1, 2) (2, 1) (2, 1) (1, 3) (1, 1) x = 0 x = 0 x = 0 x = 0 x = 0 x = 0 x = 0 x = 0

Demand 6 flow units from s to t Goal minimize time Result need 5 time units Furthermore the transit times τe could be load-dependent ⇒ NP-hardness

17 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Limits

But all these algorithms assume fixed capacities and produce unbalanced loads. On the other hand: Likelyhood of traffic collapse increases with load. Collapse → capacity drop Load is time-dependent (rush hour) So, prepare for recovery when collapse happens.

18 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Temporary collapse

Goal recover urban traffic flow after a collapse Method rerouting flow locally

19 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Probabilistic collapse networks

Previous work in Braunschweig: an algorithm to redistribute flow locally after collapse Avoids time-consuming global

• ptimization computation

When the edge has recovered return to original

• ptimization

Makes a traffic network self-healing Implement detection of collapse and its recovery with car2car communication.

20 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook State of the art Rerouting flows in reply to a collapse

Online optimization of traffic light schedules

More previous work in Braunschweig: temporary adaption of traffic light schedules to unexpected events Usually traffic light schedules are optimized offline to create “green waves” But those cannot react online on unexpected events When the traffic light detects an increased backlog the algorithm changes the schedule temporarily to react on this when situation is recovered it gets back to original

• ptimized schedule

21 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook

Next steps

Complete work on HDC aggregation and distribution Find limits of Jam-ADS and try to exceed them Implement rerouting algorithms with car2car communication Integrate HDCs and ADSs

22 11th Colloquium of SPP 1183

Review Hovering Data Clouds Improving Flow Urban traffic Outlook

Questions?

Maps ➞ by OpenStreetMap contributors, CC-BY-SA Picture ➞ by Romy1971 / PIXELIO

23 11th Colloquium of SPP 1183