A proposal for a Crowdsourcing Approach for Last Mile Delivery (CALMeD) to extend the SOURF framework. The system take advantage of the movements of citizens in urban enviroments. Application to Valencia, using its bike rental service

1. Introduction SURF MAS CALMeD Study case Conclusions
The Multi-agent layer of CALMeD SURF
M. Rebollo, A. Giret, C. Carrascosa & V. Julian
Universitat Polit`ecnica de Val`encia
D. Sistemas Inform´aticos y Computaci´on
AT 2017, ´Evry, France
c b a
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

2. Introduction SURF MAS CALMeD Study case Conclusions
Contents
1 Introduction
2 SURF Framework
3 Layer for Crowdsourcing Approach for LMD
4 Case: LMD Using Bike Sharing Service
5 Conclusions
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

6. Introduction SURF MAS CALMeD Study case Conclusions
Introduction
Main idea
Take advantage of natural movement of citizens in urban areas to
deliver goods.
apply collaborative economy last mile delivery problem (LMD)
sustainability: reduce CO2 footprint and traffic congestion
80% of deliveries at less than a 5 km distance
crowdsourcing can reduce the average trip 1.6 km
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

7. Introduction SURF MAS CALMeD Study case Conclusions
Assumptions for collaborative distribution
dynamic service demand
dynamic number of vehicles and deliverers
autonomy
scalable
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

8. Introduction SURF MAS CALMeD Study case Conclusions
SURF Framework
Deliverer_1
Deliverer_2
Deliverer_3
Deliverer_4
Deliverer_5
d
Customer
Address
Parcel
Pick-up
Address UDC
UDC
Transportation Network Analysis Module
The City and the Crowd
Potential LMD Crowd
Actual crowd
for Parcel
LMD
MAS application
Request for
Parcel LMD
CALMeD SURF
Intelligent Transportation Ontology
Running Support
SURF
New Task
Event Proccesing
Trust &
Reputation
Task
Allocation
Fleet
Tracker
fleet events
Monitoring
2
1
3 4 5
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

9. Introduction SURF MAS CALMeD Study case Conclusions
User roles scheme
we’ll focus on
individuals
being deliverer is
optional
sender: origin of
the deliver chain
giver and receiver
roles for exchanges
User
Deliverer Customer
Private Person Company
Giver Receiver Sender
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

10. Introduction SURF MAS CALMeD Study case Conclusions
Getting a LMD offer from TNAM
Deliverer CALMeD Manager
SURF
Fleet
Tracker
Monitoring
Transportation Network
Analysis Module
Deliverer Cooperation Domains: Get a LMD offer
PotentialDeliverersAndPath(ParcelID)
BuilIndividualLMD
Offer(ParcelID)
IndividualLMDOffer(LMDOfferI
D,ParcelID,ParcelInfo,
DeliveryPickUpPoint,
DeliveryDropOffPoint,
TimeWindowAtPickUpPoint,
TimeWindowAtDropOffPoint,
UserAtPickUpPoint,
UserAtDropOffPoint,
PricePerLMDService)
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

11. Introduction SURF MAS CALMeD Study case Conclusions
LMD offer negotiation
Deliverer CALMeD Manager
SURF
Deliverer Cooperation Domains: Negotiate LMD offer
In this case all the information included in the
parameters are the same as submitted by the
deliverer in its offer. In order to accept or reject it the
deliverer must execute the corresponding cooperation
domain.
Intelligent
Transportation
Ontology
Event Proccesing
Task
Allocation
IndividualLMDOfferFromDelive
rer(LMDOfferID,ParcelID,Parc
elInfo, DeliveryPickUpPoint,
DeliveryDropOffPoint,
TimeWindowAtPickUpPoint,
TimeWindowAtDropOffPoint,
UserAtPickUpPoint,
UserAtDropOffPoint,
PricePerLMDService)
CheckLMDOffer(L
MDOfferID)
IndividualLMDOffer(LMDOfferI
D,ParcelID,ParcelInfo,
DeliveryPickUpPoint,
DeliveryDropOffPoint,
TimeWindowAtPickUpPoint,
TimeWindowAtDropOffPoint,
UserAtPickUpPoint,
UserAtDropOffPoint,
PricePerLMDService)
If LMDOffer
is valid
If LMDOffer
is not valid
IndividualLMDOffer(LMDOfferI
D,ParcelID,ParcelInfo,
DeliveryPickUpPoint,
DeliveryDropOffPoint,
TimeWindowAtPickUpPoint,
TimeWindowAtDropOffPoint,
UserAtPickUpPoint,
UserAtDropOffPoint,
PricePerLMDService)
In the first case the information included in the
parameters are new values proposed by the system
and mark as counter offer. In order to accept or reject
it the deliverer must execute the corresponding
cooperation domain.
ConfirmLMDOfferCacelation(UserI
D,LMDOfferID,ParcelID)
If the
system has
a new
counter
offer
If the
system
reject the
deliverer
offer
In the second case the negotiation finishes
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

12. Introduction SURF MAS CALMeD Study case Conclusions
Parcel exchange
LMDReceiver::User
User Cooperation Domain: Start LMD
SURF
Intelligent
Transportation
Ontology
Event Proccesing
Task
Allocation
LMDGiver::User
GoToPickUpPoint(LMDOffer
ID,ParcelID,TimeToPickUp,
TransportationRoute)
Fleet
Tracker
MonitoringAtPickUpPoint(ParcelID,LMDGiverID)
AtPickUpPoint(ParcelID,LMDReceiverID)
Asynchronous
Messages
If AtPickPoint
and
OnTimeToPick
Up
GiveParcel(ParcelID,LMDGiverID,L
MDReceiverID,CurrentTimeStamp,
GPSLocation)
ScanParcel(LMDOfferID,ParcelID,LMDGiverI
D,CurrentTimeStamp,GPSLocation)
This
Message
activates
the Finish
LMD of the
Giver
GoToDeliveryPoint(LMDOffe
rID,ParcelID,TimeToPickUp,
TransportationRoute)
Synchronous
Messages
If not (AtPickPoint and
OnTimeToPickUp)
CALMeD Manager
InformLMDOfferRe-
Plan(ParcelID,Reason)
InformLMDOfferRe-
Plan(ParcelID,Reason)
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

13. Introduction SURF MAS CALMeD Study case Conclusions
Bike sharing service in Valencia
-0.44 -0.42 -0.4 -0.38 -0.36 -0.34 -0.32
39.43
39.44
39.45
39.46
39.47
39.48
39.49
39.5
39.51
Valenbisi Bike Rental Network
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

14. Introduction SURF MAS CALMeD Study case Conclusions
Network characterization
275 stations
(nodes) and 815
edges
average degree 5.9
diameter of the
network: 13 links
average shortest
path length 6.5
Posisson
distribution of
cummulative degree
distrib.
-0.44 -0.42 -0.4 -0.38 -0.36 -0.34 -0.32
39.43
39.44
39.45
39.46
39.47
39.48
39.49
39.5
39.51
Valenbisi Bike Rental Network
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

15. Introduction SURF MAS CALMeD Study case Conclusions
Network resilience. Efficiency
measure the effect
of failures in nodes
in our case, a failure
is a station with no
bikes or with no
parking places
no hubs: robust
under deliberate
failures
0 50 100 150 200 250 300
#faulty nodes
0
0.2
0.4
0.6
0.8
1
E/EG
Valenbisi Stations. Efficiency
Random
Degree
Closeness
Betweenness
Eigenvalue
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

16. Introduction SURF MAS CALMeD Study case Conclusions
User movements as L´evy flights
humans move
following L´evy
flights
short movements
with eventual long
displacements
variation: include
back to origin
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

17. Introduction SURF MAS CALMeD Study case Conclusions
User movements as L´evy flights
when the predicted
position of the final
reciver is used, L´evy
flights produce better
results
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

18. Introduction SURF MAS CALMeD Study case Conclusions
Sample trip
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

19. Introduction SURF MAS CALMeD Study case Conclusions
Static vs. dynamic positions
four sizes: 25, 50,
75 and 100
deliverers
static or in
movement final
deliverees
teams up to 50
deliverers obtain
path lengths in the
limits of network
diameter (13)
20 30 40 50 60 70 80 90 100
#nodes
10
15
20
25
30
35
40
45
50
pathlength
Average Path Length for User Delivery
static
dynamic
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF

20. Introduction SURF MAS CALMeD Study case Conclusions
Conclusions
model for crowdsourcing approach for LMD
extension for SURF framework
validation sample using the bike sharing service of Valencia
real movements have been analyzed to simulate the
displacements of users in the city and create a model
the network is robust enough to support de service
consider subjects in movement is feasible
it works better with a reduced number of deliverers
Rebollo, Giret, Carrascosa & Julian DSIC-UPV
The Multi-agent layer of CALMeD SURF