On Multi-Sensor Task Allocation
Diego PizzocaroPhD candidate
Supervisors: Prof. Alun Preece - Dr. Roger Whitaker
Seminar @ DEI University of Padova (Italy)
April, 2009
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Cardiff School of Computer Science
• 2007 - Computer Engineering, University of Padova (Italy), supervised by Prof. Luca Schenato
• 2008 - Start PhD in Computer Science at Cardiff University (UK), supervised by Prof. Alun Preece and Dr. Roger Whitaker
• International Technology Alliance (ITA) project: multi-national teams supported by complex information networks
• Our research focus: intelligent resource allocation in sensor networks.
Brief Bio
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
U.S. Army Research LabU.K. Ministry of Defence
1. Motivations & MSTA problem
2. MSTA in Homogeneous Sensor Network
3. MSTA in Heterogeneous Sensor Network
4. Taxonomy of MSTA problems
5. Conclusion
Outline
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Sensors Tasks
Simple sensors
Platforms
e.g. Search-&-Rescue mission
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Sensors Tasks
Simple sensors
Platforms
e.g. Search-&-Rescue mission
TASK 1
Injured people to identify
TASK 4
Area Surveillance
(possible threats)
TASK 3
Area Surveillance
(possible threats)
TASK 2Injured
people to identify
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Scenario
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• An already deployed network of sensors
Scenario
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• An already deployed network of sensors
- Support multiple tasks to be accomplished simultaneously
Scenario
TASK 8
Detect vehicles
TASK 7Monitor weather
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
TASK 3
Area Surveillance
TASK 5
Monitor weather
TASK 1
Injured people to identify
TASK 6
Identifyevacuation
route
TASK 2
Area Surveillance
TASK 4
Identifyevacuation
route
• An already deployed network of sensors
- Support multiple tasks to be accomplished simultaneously
Scenario
TASK 8
Detect vehicles
TASK 7Monitor weather
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
TASK 3
Area Surveillance
TASK 5
Monitor weather
TASK 1
Injured people to identify
TASK 6
Identifyevacuation
route
TASK 2
Area Surveillance
TASK 4
Identifyevacuation
route
- Sensors are scarce and in high demand.
• An already deployed network of sensors
- Support multiple tasks to be accomplished simultaneously
Scenario
TASK 8
Detect vehicles
TASK 7Monitor weather
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
TASK 3
Area Surveillance
TASK 5
Monitor weather
TASK 1
Injured people to identify
TASK 6
Identifyevacuation
route
TASK 2
Area Surveillance
TASK 4
Identifyevacuation
route
- Sensors are scarce and in high demand.
- Highly dynamic (sensor failures, change of plan)
• An already deployed network of sensors
- Support multiple tasks to be accomplished simultaneously
Scenario
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
TASK 1
Injured people to identify
TASK 2
Area Surveillance
- Sensors are scarce and in high demand.
- Highly dynamic (sensor failures, change of plan)
“Where is it better to send that particular UAV?”
• We need schemes to allocate sensors to the task they best serve, considering all the relevant parameters.
• In general we can have
• static or mobile sensing devices
• tasks requiring multiple sensors or one sensor
• sensors shared or not shared between multiple tasks
• etc...
MSTA problem
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• We need schemes to allocate sensors to the task they best serve, considering all the relevant parameters.
• In general we can have
• static or mobile sensing devices
• tasks requiring multiple sensors or one sensor
• sensors shared or not shared between multiple tasks
• etc...
MSTA problem
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Which sensor should be allocated to which task?
• The fundamental question remains:
• We need schemes to allocate sensors to the task they best serve, considering all the relevant parameters.
• In general we can have
• static or mobile sensing devices
• tasks requiring multiple sensors or one sensor
• sensors shared or not shared between multiple tasks
• etc...
MSTA problem
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Which sensor should be allocated to which task?
Multi-Sensor Task Allocation (MSTA)
• The fundamental question remains:
• MSTA arises in a variety of domains:
• environmental monitoring
• natural disaster (e.g. earthquakes), ....
MSTA applications
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• MSTA arises in a variety of domains:
• environmental monitoring
• natural disaster (e.g. earthquakes), ....
• We focus on military/humanitarian scenarios
‣ our allocation mechanisms can be applied to other domains!
MSTA applications
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• MSTA arises in a variety of domains:
• environmental monitoring
• natural disaster (e.g. earthquakes), ....
• We focus on military/humanitarian scenarios
‣ our allocation mechanisms can be applied to other domains!
• The task allocation process differs a lot when
‣ in a homogeneous sensor network (e.g. only seismic sensors)
‣ in a heterogeneous sensor network (e.g. seismic+UAVs+UGVs)
MSTA applications
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• MSTA arises in a variety of domains:
• environmental monitoring
• natural disaster (e.g. earthquakes), ....
• We focus on military/humanitarian scenarios
‣ our allocation mechanisms can be applied to other domains!
• The task allocation process differs a lot when
‣ in a homogeneous sensor network (e.g. only seismic sensors)
‣ in a heterogeneous sensor network (e.g. seismic+UAVs+UGVs)
• We discuss two examples of MSTA instances in both networks
MSTA applications
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
MSTA in HomogeneousSensor Network
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• Governmental institutions (ARL, MoD), and researchers suggest Wireless Sensor Networks (WSNs) as “the future” for military operations.
Why Homogeneous SN ?
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• Governmental institutions (ARL, MoD), and researchers suggest Wireless Sensor Networks (WSNs) as “the future” for military operations.
• WSNs are often composed exclusively by hundreds of cheap miniaturized wireless sensors (called motes) with the same sensing capabilities.
Why Homogeneous SN ?
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• Governmental institutions (ARL, MoD), and researchers suggest Wireless Sensor Networks (WSNs) as “the future” for military operations.
• WSNs are often composed exclusively by hundreds of cheap miniaturized wireless sensors (called motes) with the same sensing capabilities.
• In general every network composed exclusively sensors with the same sensing capabilities is called “Homogeneous Sensor Network”.
Why Homogeneous SN ?
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• We considered a particular instance of the MSTA problem in a Homogeneous Sensor Network (see DCOSS 08).
• Assumptions:
‣ Sensors can serve only one task per time: Single-Task sensors
- Therefore tasks are competing for the exclusive usage of a sensor
‣ A task might require more than one sensor: Multi-Sensor tasks
‣ Available info does not permit planning for future: Instantaneous allocation
Problem settings
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
T2
T1
x
T1
• Tasks:
‣ vary in priority
‣ have a different demand for sensing resource capabilities
‣ have to respect a budget (e.g. monetary).
• Each sensor:
‣ has a different utility for each task(e.g. geography & distance)
‣ has a different cost for each task.
• Goal: maximizes the utilitywhile not exceeding the budgets of each task.
Formal model
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
S1
S2
S3
S4
T1
T2
Sensors
Tasks(e11, c11)
(p1, d1, b1)(e
12, c12)
(p2, d2, b2)
e = utility of sensor to a task
c = cost of a sensor to a task
p = task priority
d = task utility demand
b = task budget
• At least as hard as the Knapsack problem which is NP-Complete
➡ we developed heuristic algs and compared them with state of the art pre-existent approaches.
• Simulation environment implemented in Java (with Penn State University)
Allocation algorithms
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• At least as hard as the Knapsack problem which is NP-Complete
➡ we developed heuristic algs and compared them with state of the art pre-existent approaches.
• Simulation environment implemented in Java (with Penn State University)
• The algorithm which offers the best trade-off optimality Vs computational cost: MRGAP algorithm:
‣ a centralized algorithm: i.e. we collect all the info about the network in a single node
‣ it can be easily implemented as a distributed algorithm:less communication overhead.
Allocation algorithms
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
MSTA in HeterogeneousSensor Network
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• Sensor network deployment during military/humanitarian missions:
‣ UAV will fly on the battlefield and drop hundreds of motes
‣ Some motes might also be mobile.
Why Heterogeneous SN ?
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• Sensor network deployment during military/humanitarian missions:
‣ UAV will fly on the battlefield and drop hundreds of motes
‣ Some motes might also be mobile.
• Therefore we have a sensor network composed by sensors with different sensing capabilities and mobility capabilities: “Heterogeneous sensor network”
Why Heterogeneous SN ?
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• Sensor network deployment during military/humanitarian missions:
‣ UAV will fly on the battlefield and drop hundreds of motes
‣ Some motes might also be mobile.
• Therefore we have a sensor network composed by sensors with different sensing capabilities and mobility capabilities: “Heterogeneous sensor network”
• Another example: Sensor Web by Open Geospatial Consortium (OGC) with environmental monitoring applications.
Why Heterogeneous SN ?
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• We considered a particular instance of the MSTA problem in a Heterogeneous Sensor Network.
• Same assumptions:
‣ Single-Task sensors
‣ Multi-Sensor tasks
‣ Instantaneous allocation
• Difference with homogeneous case:
‣ Combined utilities of groups of sensors (bundles) are in general much complex to compute than the homogeneous SN.
Problem settings
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
T2
T1
x
T1
TASK 1
Area video Surveillance
(possible threats)
• We considered a particular instance of the MSTA problem in a Heterogeneous Sensor Network.
• Same assumptions:
‣ Single-Task sensors
‣ Multi-Sensor tasks
‣ Instantaneous allocation
• Difference with homogeneous case:
‣ Combined utilities of groups of sensors (bundles) are in general much complex to compute than the homogeneous SN.
Problem settings
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
T2
T1
x
T1
TASK 1
Area video Surveillance
(possible threats)
Sensor bundles
Formal model
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
S1
S2
S3
S4
B1
B2
SensorsBundles
e = joint utility of a bundle to a taskp = task priority
T1
T2
Tasks
(p1)
(p2)
e11
e12
• We first want to group sensors into bundles, and then we want to find the best assignment of bundles to tasks.
• Problem well studied in Multi-agent Systems: Coalition formation
• Typical approach: combinatorial auction
‣ bidders: tasks
‣ items: sensors
‣ tasks bids for bundles of sensors
Allocation mechanisms (1)
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• Problem well studied in Multi-agent Systems: Coalition formation
• Typical approach: combinatorial auction
‣ bidders: tasks
‣ items: sensors
‣ tasks bids for bundles of sensors
Allocation mechanisms (1)
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• Need to enumerate all possible bundles for each task.
• Large number of sensors and tasks: the computational cost is too large.
• Our contribution:Prune the set of bids placed by tasks (i.e. reduce the number of possible bundles).
• We define a system architecture in which we gradually reduce the search space of the allocation algorithms (see EKAW08).
Allocation mechanisms (2)
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• We define a system architecture in which we gradually reduce the search space of the allocation algorithms (see EKAW08).
• Three main components:
Allocation mechanisms (2)
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Reasoner
Bundle
Generator
AllocationAlgorithms
< Package Config >
{ < Bundle1, e1 >,
< Bundle2, e2 >, ... }
Sensor types compatible with the task.
Sensor Bundles generated based on the package configuration.
• We define a system architecture in which we gradually reduce the search space of the allocation algorithms (see EKAW08).
• Three main components:
• Current work:Implementing/Testing this approach using the simulation environment “Player/Stage”.
Allocation mechanisms (2)
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Reasoner
Bundle
Generator
AllocationAlgorithms
< Package Config >
{ < Bundle1, e1 >,
< Bundle2, e2 >, ... }
Sensor types compatible with the task.
Sensor Bundles generated based on the package configuration.
Taxonomy of MSTA problems
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• MSTA is closely related to Multi-Robot Task Allocation (MRTA):
“Which robot should execute which task?"
in a Multi-Robot System (MRS).
Related work - MRTA
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• MSTA is closely related to Multi-Robot Task Allocation (MRTA):
“Which robot should execute which task?"
in a Multi-Robot System (MRS).
• Gerkey et al (2004) proposed an MRTA taxonomy:
➡ MRTA problems can be viewed as instances of other well-studied,
optimization problems.
➡ therefore allowing comparison of different solutions.
Related work - MRTA
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• We propose a preliminary MSTA taxonomy as an extension of MRTA to cover important features of sensor networks (INFOCOM 09).
Preliminary MSTA taxonomy
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• We propose a preliminary MSTA taxonomy as an extension of MRTA to cover important features of sensor networks (INFOCOM 09).
Preliminary MSTA taxonomy
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• MSTA taxonomy organized on four main axes:
1. Sensors: Single-task (ST) vs. multi-task (MT).
2. Tasks: Single-sensor (SS) vs. multi-sensor (MS).
3. Assignment: Instantaneous (IA) vs. time-extended (TA).
4. Sensor Network: Homogeneous (HO) vs. heterogeneous (HE).
• Example: previously we have considered ST-MS-IA-HO and ST-MS-IA-HE
• We have presented the general MSTA problem,
• Discussed two MSTA instances with applications in military/humanitarian missions,
• Outlined the need for a MSTA taxonomy and presented a preliminary version of it.
Conclusion
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
• We have presented the general MSTA problem,
• Discussed two MSTA instances with applications in military/humanitarian missions,
• Outlined the need for a MSTA taxonomy and presented a preliminary version of it.
• Future research:
‣ Refine MSTA taxonomy
‣ What are the most important SN features to include?
‣ Explore different MSTA instances:
‣ Heterogeneous SN with Multi-Task sensors (sensors can be shared)
Conclusion
http://users.cs.cf.ac.uk/D.Pizzocaro [email protected]
Thanks for listening !
Cardiff School of Computer Science
http://users.cs.cf.ac.uk/D.Pizzocaro