Vehicle Routing in Practice - SINTEF · The Vehicle Routing Problem (VRP) Given a fleet of vehicles...
Transcript of Vehicle Routing in Practice - SINTEF · The Vehicle Routing Problem (VRP) Given a fleet of vehicles...
-
1ICT
Vehicle Routing in Practice
Geir HasleSINTEF ICT, Oslo, Norway
XVIII EWGLANaples, Italy April 28-30 2010
-
2ICT
Outline
Background and ContextThe Vehicle Routing ProblemIndustrial AspectsSpider - A Generic VRP SolverSpecific casesOngoing and Further Research
-
3ICT
Technology for a better society
-
4ICT
Norway
Oslo
Trondheim
An Independent Multi-DisciplinaryContract R&D OrganisationEstablished in 1950Among the Largest CRO in Europe
Vision:Technology for a better Society
Business Concept:To meet the needs for Research-Based Innovation and Development for the Private and Public Sectors
450 employees in Oslo
1.300 employees in Trondheim
-
5ICT
SINTEF - A Norwegian Contract Research InstituteScience and engineering – social sciences, health careConnected with
The Norwegian University of Science and Technology, TrondheimThe University of Oslo, Faculty of mathematics and natural sciences
> 90 % of turnover from industry and public sector contractsAnnual turnover ~ 200 M€Activities
Strategic, long term, basic researchContract ResearchConsultancyCommercialization and spin-offs
-
6ICT
The SINTEF GroupRe
sear
ch D
ivisio
ns
SINTEFHealth Research
SINTEFTechnology and
SocietySINTEF
ICTSINTEF
Materials and Chemistry
SINTEF Building and Infrastructure
SINTEF Petroleum and Energy
SINTEF Petroleum ResearchSINTEF Energy Research
SINTEFHolding
PresidentExecutive
Vice President
SINTEF’s CouncilSINTEF’s Board
Corporate Staff
SINTEF MarineMarintek
SINTEF Fisheries and Aquaculture
-
7ICT
Outline
Background and ContextThe Vehicle Routing ProblemIndustrial AspectsSPIDER - A Generic VRP SolverSpecific casesOngoing and Further Research
-
8ICT
The Vehicle Routing Problem (VRP)
Given a fleet of vehicles and a set of transportation orders, find a minimum cost routing plan. That is: allocate each order to a
vehicle, and for each vehicle, sequence the stops.
The VRP central to efficient transportation managementApplications
Distribution or pick-up of goodsDial-a-rideMunicipal servicesRepairman problemNewspaper distributionWaste managementGritting, snow clearing
Very hard, discrete optimization problem
-
9ICT
The Capacitated VRP (CVRP)Graph G=(N,A)
N={0,…,n} Nodes0 Depot, i≠0 Customers A={(i,j): i,j∈N} Arcscij >0 Arc cost
Demand di for each Customer ieither all pickups or all deliveries
V set of (identical) Vehicles, each with Capacity qGoal
Construct a set of minimum cost Routes that start and finish at the Depot.Each Customer shall be visited once (no order splitting)Total Demand for all Customers serviced less than Capacity qCost: Total Arc cost (also # Routes in richer VRP variants)
DVRP – constraint on route lengthVRPTW – VRP with time windowsVRPB – VRP with backhauling (deliveries first, then pickups)PDP – pickup and delivery without going through the depot
I aI b
I c
I d
I eI f
I g
I hIi
Ij
2
3Ik
IlImInIo
Ip 41
IlIm
IkInIo
IpI e
I d
I c
I aI bI g
I hIi
IjI f1
2
3
4
-
10ICT
Mathematical formulation of VRPTW (vehicle flow formulation)
( , )
0
, 1
minimize (1)
subject to:1, (2)
, (3)
1, (4)
0, , (5)
1, (6)
( ) 0, ( , ) , (7)
kij ij
k V i j A
kij
k V j Nk
i iji C j N
kj
j Nk kih hj
i N j Nki n
i Nk k kij i ij j
i
c x
x i C
d x q k V
x k V
x x h C k V
x k V
x s t s i j A k Va
∈ ∈
∈ ∈
∈ ∈
∈
∈ ∈
+∈
= ∀ ∈
≤ ∀ ∈
= ∀ ∈
− = ∀ ∈ ∀ ∈
= ∀ ∈
+ − ≤ ∀ ∈ ∀ ∈
∑ ∑
∑∑
∑ ∑
∑
∑ ∑
∑
, , (8){0,1}, ( , ) , (9)
ki i
kij
s b i N k Vx i j A k V≤ ≤ ∀ ∈ ∀ ∈∈ ∀ ∈ ∀ ∈
minimize cost
each customer once
capacity
k routes out of depot
flow balance for each customer
k routes into depot (redundant)
start of service and driving time
start of service within TWarc (i,j) driven by k
-
11ICT
The VRP is very hard, computationally
Belongs to class of strongly NP-hard optimization problemsComputing time for all VRP optimization algorithms grows”exponentially” with the number of nodes (probably ... )Exact methods are ”easy” to design, may take foreverCurrently, exact methods for the C/DVRP can consistentlysolve instances with some 70 customers only to optimalitywithin a few minutesFor generic industrial VRP tools, we must give up the questfor optimality and resort to some form of approximationWhat is the optimal plan, anyway?
-
12ICT
VRP in Operations Research
since 1959thousands of papersmostly generic work on idealized VRP modelsstill, one of the successes of ORa tool industry has emerged, based on research resultstremendous increase in ability to ”solve” VRP variants
general increase of available computing powermethodological improvement
still much to do, VRP research has never been more activeshort road from research results to industrial benefits
-
13ICT
VRP in Operations ResearchGeneral, idealized formulationsExtensions studied in isolation
time windowsmultiple depotsinventory constraints....
Very fruitful for understanding VRP variantsdeveloping highly targeted algorithms
Not always relevant to real-life problemsSome application specific workRecently: More holistic approach: ”Rich VRPs”
General model, many aspects of industrial problemsRobust algorithms
-
14ICT
Extensions in the VRP-literatureLocation Routing LRPFleet Size and Mix FSMVRPVRP With Time Windows VRPTWGeneral Pickup and Delivery GPDPDial-A-Ride DARPPeriodic VRP PVRPInventory Routing IRPDynamic VRP DVRPCapacitated Arc Routing Problem CARP
All kinds of algorithmic approachesexactapproximation methods
based on systematic, exact methodsmetaheuristics
-
15ICT
Outline
Background and ContextThe Vehicle Routing ProblemIndustrial AspectsSpider - A Generic VRP SolverSpecific casesOngoing and Further Research
-
16ICT
Industrial aspects – VRP toolAdequate model of the applicationsWide range of applicationsHigh quality information
addressesdistances, times, costsordersfleet and driver information
Manual planning, user interfaceEase of useSoftware issues
Integration with ERP etc.ExtendabilityMaintainabilityDocumentation
High quality solution in short time – powerful VRP solver
-
17ICT
Industry needs rich VRP modelsType of service and operation
multiple depots, no depotdifferent order types (delivery, pickup, direct, service, ...)order splitting, flexible order volumesnode routing, arc routing, mixedmultiple tours per dayperiodic problemsconnection to inventory and manufacturing
Constraintscapacity, several dimensions, hard and softtime windows, multiple, hard and softprecedences(in)compatibilities
Realistic distances, times, costs Cost components
multiple criteriasoft constraints and penalties
Uncertainty and Dynamics
-
18ICT
Research on Rich VRPs & related problems at SINTEFIndustrial Contracts since 1995Strategic Research
European Commission FP III, IV, V (e.g., GreenTrip 1996-1999)Norwegian Research Council (RCN)Internal Projects, students
Generic VRP Solver - Spider (1995→)Commercialization from 1999GreenTrip AS → Spider Solutions AS
TOP Programme 2001-2004 (RCN) http://www.top.sintef.no/Basic Research on Rich VRP and related problemsVRPTWShortest Path Problem in Dynamic Road Topologies
Innovation Projects supported by Research Council of Norway“I Rute” (2001 – 2004) Bulk transportation“DOiT” (2004 – 2007) Stochastic and Dynamic Routing“EDGE” (2005 – 2008) Large Scale Routing“Effekt” (2008 – 2011) Media Product Distribution Routing
Ship routingTurboRouter 1995 ->SINTEF Internal Strategic Project (2005-2008)LNGShipping
http://www.top.sintef.no/
-
ProductionProduction portport
ConsumptionConsumption portport
ProductProduct 11
ProductProduct 22 KjKjøøpsvikpsvik
BrevikBrevik
AltaAlta
Mo i RanaMo i Rana
TrondheimTrondheim
KarmKarmøøyy
ÅÅrdalrdal
-
20ICT
Invent – solving maritime routing problems
Invent is a software library that can model and solve a wide class of maritime routing problems. This includes both traditional tramp shipping and industrial shipping problems with inventory management.
Vessels can be modeled at tank level with the possibility of detailed tank stowage and cleaning operations.
Inventories can have a time varying production or consumption profile together with an upper and lower limit on the inventory level.
The inventories can be combined with traditional orders with laycan and quantity intervals for a pair of pickup and delivery locations.
It is also possible to include contracts with details on required pickup or delivery in given periods, and time varying price curves for income and cost.
Problem instances are described in a general XML format that can be used as input for the automatic planning process.
Invent generates optimized plans for the problems based on advanced heuristic methods that are able to provide high quality solutions in short time.
Invent
xmlC++ API
User application
-
21ICT
TurboRouter
-
22ICT
Maritime Transport Optimization AN OCEAN OF OPPORTUNITIES
ORMS Today, Special International Issue, Vol 36 No 2 pp 28-31, April 2009.
http://viewer.zmags.com/publication/1368b369#/1368b369/28
http://viewer.zmags.com/publication/1368b369#/1368b369/28
-
23ICT
Outline
Background and ContextThe Vehicle Routing ProblemIndustrial AspectsSpider - A Generic VRP SolverSpecific casesOngoing and Further Research
-
24ICT
Spider - A Generic VRP Solver
Designed to be widely applicableBased on generic, rich modelPredictive route planningPlan repair, reactive planningDynamic planning with stochastic model
Framework for VRP research
-
25ICT
SPIDER - Generalisations of CVRPHeterogeneous fleet
CapacitiesEquipmentArbitrary tour start/end locationsTime windowsCost structure
Linked tours with precedencesMixture of order typesMultiple time windows, soft time windowsCapacity in multiple dimensions, soft capacityAlternative locations, tours and ordersArc locations, for arc routing, mixed problems and aggregation of node ordersAlternative time periodsNon-Euclidean, asymmetric, time-varying travel timesCompatibility constraintsA variety of cost components and soft constraints
driving time restrictionsvisual beauty of routing plan, non-overlapping routeslevelling
-
26ICT
Transportation OrderDifferent types:
DeliveryPickupDirect (P&D)Service
Plan structuretasktask sequencetourplan
-
27ICT
Tour
Time windows and locations given by start/stop tasksSelected vehicle and driver (alternative equipages)Linked tours
(may have different vehicles)
-
28ICT
Vehicle and Driver
CapacityTravelling attributes
Speed profileHeight, weight, lengthObey one-way restrictions?
Driving time regulations
-
29ICT
Cost elements
Travel cost (distance, time, tolls)Tour usage cost
Cost for starting a tourCost per order on tour
Cost for unserviced orderWaiting time costCost for alternative locationsCost for same/different locationCost for breaking work regulationsCost for “ugly”, overlapping routes
-
30ICT
Constraints
ConsistencyComplete orderPickup/delivery (Direct orders) same tour and precedence
TimeTravel time, DurationTime windows, multiple, hard and soft
Vehicle capacities, multiple dimensions, hard and softTotal capacities over a set of tours
-
31ICT
Constraints
Compatibility vehicle/order vehicle/location product/compartmentOrders on same tourCorresponding locations (when alternatives)
Order: Choose corresponding locations for pickup and delivery taskTour: Choose corresponding locations for start/stop tasks
Corresponding time periods for sets of ordersE.g. Delivery day 1,3,5 or day 2,4,6
-
32ICT
Locks
Prevent optimiser changing part of planTask: Time lockTour: Lock whole or initial part of tourOrder: Lock (un)assignment
-
33ICT
Uniform Algorithmic ApproachGoals
Reach a good local optimum fastExplore interesting parts of search space efficiently
3 phasesConstructionIterative Improvement: Iterated Variable Neighborhood DescentTour Depletion
Based onIterated Local Search (Martin, Lourenço et al)Variable Neighborhood Descent (Hansen & Mladenovic)Diversification when VND reaches local optimum
-
34ICT
Construction of Initial Solution
Various Sequential Construction HeuristicsExtended to cover Richer Model
Several types of orderNon-standard constraintsNon-homogeneous fleetMultiple DepotsMultiple Tours per Vehicle
New ConstructorInhomogeneous problemsMultiple depotsMultiple tours per vehicleMore search
-
35ICT
Variable Neighborhood Descent - Operators
InsertRemove2-optOr-opt3-optRelocateCross (2-opt*)Exchange (Cross-Exchange)Tour DepletionChange alternative locationChange alternative time periodChange vehicle...
Variety, efficiency, sequence?
-
36ICT
Illustration: ExchangeInter-tour operatorFull neighborhood is typically largeRemedy: Limit on maximum segment lengthAlternative: Focus on promising cut points
-
37ICT
Iterated Local Search (Martin, Lourenço et al)
Goal: Efficient search in new basins of attractionWhen VND reaches local optimum: diversifyRandom restartAlternative initial solutionPath relinkingNoising: perturb problem dataChange objectiveLNS, VLNS
-
38ICT
Take away a substantial number of commitmentsrandomly“similar” commitments
Reconstruct with fast insertion methodCheapest insertionRegret-based insertion
Accept new solution ifbetter, diverseThreshold AcceptanceSimulated Annealing
IterateAlternative modifications
Limited Local SearchFull VND MachineryDistance Metrics
Large Neighborhood Search (Shaw)
-
39ICT
Computational Experiments
Standard test problems in literature for most VRP variants”World championship” in Vehicle RoutingFor most instances, the optimal solution is not knownTesting against the best methods reported in the literature”Unfair” comparison: competing with solvers optimized for solving a specific, idealized problemNice to know how well you perform ...
-
40ICT
Experimental studies PDPTW http://www.sintef.no/top
Author 100 200 400 600 800 1000 TotalLi & Lim 41 14 6 2 0 0 63Spider 13 11 5 4 9 4 46BVH 2 8 6 3 4 1 24TS 0 1 2 0 0 2 5SR 0 26 41 51 47 51 216
Total 56 60 60 60 60 58 354
354 standard test problems Li & Lim: 100 – 1.000 orders
BVH: Bent & Van Hentenryck: Two-stage, Hybrid Local Search1. Minimize # tours by SA, modified objective2. Minimize Distance with Large Neighborhood Search
TS: Tetrasoft, Danish tool vendor, Method unknownSR: Stefan Röpke, Univ. Copenhagen
Adaptive Large Neighborhood Search
http://www.sintef.no/top
-
41ICT
Results on CVRP - Rounded distances (1)
Augerat et al 74 instances, 15-100 customers, limit on # vehicles A (27): 0.15% from optimum, 19 optimal, 0-541 s, average 100 sB (23): 0.11% from optimum, 17 optimal, 1-582 s, average 95 sP (24): 0.23% from optimum, 15 optimal, 0-450 s, average 99 s
Christofides & Eilon (1969)13 instances, 12-100 customers, limit on # vehiclesE (13): 0.32% from optimum, 2 optimal, 7 - 487 s, average 178 s
Fisher3 instances, 44 - 134 customers, limit on # vehiclesF (3): 0.27% from optimum, 7 optimal, 0-289 s, average 103 s
Gillett & Johnson, 1 instance: G-n262-k25best known, -7.09% from previous best known (5685)
-
42ICT
G-n262-k25: 5685 vs. 6119
-
43ICT
Results on CVRP - Rounded distances (2)
Christofides, Mingozzi & Toth (1979)5 CVRP instances, 100-199 customers, limit on # vehicles0.99 % from optimum, 21-1255 s, average 404s1 optimal1 new best known -1.6 %1 worse by 2.7 %1 worse by 2.9 %M-n200-k16: no feasible solution previously knownFeasible solution found
after 533s: distance 1371
-
44ICT
M-n200-k16: First known feasible solution
0
10
20
30
40
50
60
70
80
0 10 20 30 40 50 60 70
-
45ICT
Computational tests - NEARP
Prins & Bouchenoua CBMix (23 instances)No lower bounds, no proven optima, only one competitorUB error 0.94%8 best known solutions (6 new)
-
46ICT
Comparison with Prins & Bouchenoua
-2.00
-1.00
0.00
1.00
2.00
3.00
4.00
5.00
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
%Error
-
47ICT
Products - architecture
Guider(topology)
Planner(VRP solver, COM)
Designer
SPIDER Solutions AS
Distribution Innovation AS
Geomatikk AS
Web-server (Servlet, C++/Java)
Web solution
Guider(topology)
Spider Server
Guider(topology)
Planner(VRP solver, COM)
DI web solution
-
48ICT
-
49ICT
Spider Designer - ApplicationsDistribution of bread (Bakers)Mail collection and distribution (Posten Norge)Local pickup and delivery (Schenker)Newspaper distribution, 1st tier (Aftenposten, Dagbladet)Newspaper distribution, last mile (Aftenposten, Stavanger Aftenblad)Collection of milk from farms (TINE)Distribution of fodder to farms (Landbruksdistribusjon)Distribution of fuel oil (Hydro Texaco)Location analyses, depot (obnoxious facility location, Norsk Gjenvinning AS)Distribution of blood (Ullevål sykehus)Distribution of groceries (REMA 1000)Distribution of magazines (Bladcentralen)Distribution of ice cream (Diplom Is; Hennig Olsen)Dial-a-ride, elderly, hospital patients (Nor-Link)
Savings 2-35%, depending on application
-
50ICT
Challenges for Routing Technology
Industrial awarenessRange of applicationsInformation availability and qualityUser interfacesModel adequacy and flexibilitySoftware engineeringRobustness of solution methodSolution quality for large-size and complex problems
-
51ICT
Challenges for Spider
Electronic road network - GISInstance robustness
all kinds of instancesScalability
very large size problemsExtendability
new operators, new VND sequence
Exploiting modern commodity computersmulti-core, heterogeneous, GPUs
-
52ICT
Outline
Background and ContextThe Vehicle Routing ProblemIndustrial AspectsSpider - A Generic VRP SolverSpecific casesOngoing and Further Research
-
53ICT
Scalability – Solving ”Huge” VRPsFaster VRP solver
heuristic investigation of neighborhoodsclever sequencing of constraint and objective component evaluationmore powerful metaheuristicshybrid methods, even combining exact methods and heuristics
Problem Decompositiongeographicallytemporallybased on productclustering methods will be usefulsubproblem
Abstraction and Aggregationclustering methods will be useful
Exploit modern parallel and heterogeneous computer architecturesMulti-level search and collaborative solvers
-
54ICT
Outline
Background and ContextThe Vehicle Routing ProblemIndustrial AspectsSpider - A Generic VRP SolverSpecific casesOngoing and Further Research
-
55ICT
Products - architecture
Guider(topology)
Planner(VRP solver, COM)
Designer
SPIDER Solutions AS
Distribution Innovation AS
Geomatikk AS
Web-server (Servlet, C++/Java)
Web solution
Guider(topology)
Spider Server
Guider(topology)
Planner(VRP solver, COM)
DI web solution
-
56ICT
NewspaperdistributionCity of Oslo500k inhabitants200k households35k modules
-
57ICT
Vehicle Routing in Practice
-
58ICT
-
59ICT
-
60ICT
-
61ICT
-
62ICT
-
63ICT
-
64ICT
-
65ICT
-
66ICT
-
67ICT
Problem description
Last mile part of two-echelon distribution(Open) DVRP with extensionsObjectives
total durationroute balancingclustering, route separation, ”visual beauty”
Constraintsroute duration# routes
-
68ICT
Additional niceties
Determination of vehicle type (pedestrian, car)Combined routesLinks that may be used by one route only...
-
69ICT
Extended problem
Integrated problemDistribution from print shop to subscriberLocation routing: Determination of pickup points
-
70ICT
Demand aggregation based on road topology, proximity
Distance/time, capacity thresholdIssues on traversal possibilities, constraintsTypical reduction factor of 5-20Needs extention to arc model (Node Edge Arc Routing Problem, NEARP)
-
71ICT
Case: Oslo data from Aftenposten33.200 orders reduced to 5600 aggregates
-
72ICT
-
73ICT
Decomposition approaches
GeographicalCluster-first route second
Capacitated Clustering Problem – with balancingRouting
-
74ICT
Outline
Background and ContextThe Vehicle Routing ProblemIndustrial AspectsSpider - A Generic VRP SolverSpecific casesOngoing and Further Research
-
75ICT
Important trends in VRP research
Richer models, larger instancesExact methodsSelf-adaptationHybrid and collaborative methodsParallel and heterogeneous computing
-
76ICT
Parallel computing Some tasks in EA and LS are embarassingly parallel”Simple” parallelization through multi-threading
fine-grained to medium level granularityvery interesting for routing technologynot so interesting for VRP research, no literature
More interesting parallelizationCoarse-grained, asynchronousMulti-searchCollaborative searchParallel multi-level
Recent review by Crainic, 80 referencesAdditional 30 papersHeterogeneous computing not really investigated
-
77ICT
Parallel and heterogeneous computing
CPU Clock frequency has stagnated(The Beach Law does not hold anymore ...)Moore’s Law still validMulti-core and heterogeneous commodity computersFine-grained and coarse-grained parallelism
-
78ICT
-
79ICT
-
80ICT
-
81ICT
The Collab project
”Iteration level” parallelization of Spider Experimental VRP Solverhttp://www.sintef.no/Projectweb/Collab/
Special session ”Metaheuristics on graphics hardware”under the META’10 conference in Djerba Island, October28-30 2010, deadline May 15 http://www2.lifl.fr/META10/
http://www.sintef.no/Projectweb/Collab/http://www2.lifl.fr/META10/index.php?n=Main.InfoMGHhttp://www2.lifl.fr/META10/
-
82ICT
SummaryThe Vehicle Routing Problem is a key to efficient transportationThe VRP is a very hard optimization problemThousands of VRP papers, mostly generic, idealizedOR success story, tool industryIndustry needs rich models, powerful algorithmsStill challenges, VRP research has never been more activeGeneric VRP Solver SPIDER
rich, generic modelresolution algorithm based on metaheuristicsresults
Still a lot of work to be done ...
-
83ICT
Geir Hasle and Oddvar Kloster: Industrial Vehicle Routing.Chapter (pp 397-435) in
Hasle, Geir; Lie, Knut-Andreas; Quak, Ewald (Eds.) Geometric Modelling, Numerical Simulation, and Optimization:
Applied Mathematics at SINTEF2007, XI, 558 p. 162 illus., 59 in color., Hardcover. ISBN: 978-3-540-68782-5
http://www.springer.com/
-
84ICT
Special Issue Transportation Science
Advances in Vehicle RoutingTRISTAN VII, Tromsø, Norway June 20-25 2010Guest editors
Marielle ChristiansenArne LøkketangenGeir Hasle
Deadline October 15, 2010
See May issue of Transportation Science for call
-
85ICT
Vehicle Routing in Practice
Geir HasleSINTEF ICT, Oslo, Norway
XVIII EWGLANaples, Italy April 28-30 2010
Vehicle Routing in PracticeOutlineSlide Number 3Slide Number 4SINTEF - A Norwegian Contract Research Institute�The SINTEF GroupOutlineThe Vehicle Routing Problem (VRP)The Capacitated VRP (CVRP)Mathematical formulation of VRPTW�(vehicle flow formulation)The VRP is very hard, computationallyVRP in Operations ResearchVRP in Operations ResearchExtensions in the VRP-literatureOutlineIndustrial aspects – VRP toolIndustry needs rich VRP modelsResearch on Rich VRPs & related problems at SINTEF Slide Number 19Invent –�solving maritime routing problemsTurboRouterMaritime Transport Optimization�AN OCEAN OF OPPORTUNITIESOutlineSpider - A Generic VRP SolverSPIDER - Generalisations of CVRPTransportation OrderTourVehicle and DriverCost elementsConstraintsConstraintsLocksUniform Algorithmic ApproachConstruction of Initial SolutionVariable Neighborhood Descent - OperatorsIllustration: ExchangeIterated Local Search �(Martin, Lourenço et al)Slide Number 38Computational ExperimentsExperimental studies PDPTW�http://www.sintef.no/topResults on CVRP�- Rounded distances (1)Slide Number 42Results on CVRP�- Rounded distances (2)Slide Number 44Computational tests - NEARPSlide Number 46Products - architectureSlide Number 48Spider Designer - ApplicationsChallenges for Routing TechnologyChallenges for SpiderOutlineScalability – Solving ”Huge” VRPsOutlineProducts - architectureSlide Number 56Vehicle Routing in PracticeSlide Number 58Slide Number 59Slide Number 60Slide Number 61Slide Number 62Slide Number 63Slide Number 64Slide Number 65Slide Number 66Problem descriptionAdditional nicetiesExtended problemDemand aggregation based on road topology, proximityCase: Oslo data from AftenpostenSlide Number 72Decomposition approachesOutlineImportant trends in VRP researchParallel computing Parallel and heterogeneous computingSlide Number 78Slide Number 79Slide Number 80The Collab projectSummarySlide Number 83Special Issue Transportation ScienceVehicle Routing in Practice