Agent-based sensor-mission assignment for tasks sharing
assetsThao Le
Timothy J NormanWambertoVasconcelos
www.usukita.orgwww.csd.abdn.ac.uk/research/ita
Structure
• Introduction & Motivation• Problem description• MSM & GAP-E• Experimental results• Discussion• Conclusion
Introduction & Motivation WSNs consist of a
large number of sensing resources
Introduction & Motivation WSNs consist of a
large number of sensing resources
form an ad-hoc network communicating with
each other and with data processing centres using wireless links
Introduction & Motivation
WSNs are required to support multiple missions
arriving at anytime decomposing into many
tasks
Introduction & Motivation
WSNs are required to support multiple missions
arriving at anytime decomposing into many
tasks may occur
simultaneously
Introduction & Motivation
WSNs are highly dynamic in terms of:
configuration: sensors move out of range or be damaged, changing weather conditions may interfere with communication, etc...
the environment: missions and phenomena occur frequently and simultaneously
The problem: Sensor-Mission Allocation
Introduction & Motivation
Motivations: to be more applicable in
realistic environmentsheterogeneous sensors & tasks
Introduction & Motivation
Motivations: to be more applicable in
realistic environments:heterogeneous sensors & tasks
to save limited energy of sensor resources in real-world applicationallowing sensors to be shared between multiple tasks
Introduction & Motivation
Motivations: to be more applicable in
realistic environments:heterogeneous sensors & tasks
to save limited energy of sensor resources in real-world applicationallowing sensors to be shared between multiple tasks
Introduction & Motivation
Motivations: to be more applicable in
realistic environments:heterogeneous sensors & tasks
to save limited energy of sensor resources in real-world applicationallowing sensors to be shared between multiple tasks
Introduction & Motivation
Motivations: to be more applicable in
realistic environments:heterogeneous sensors & tasks
to save limited energy of sensor resources in real-world applicationallowing sensors to be shared between multiple tasks
Introduction & Motivation
Motivations: to be more applicable in
realistic environments:heterogeneous sensors & tasks
to save limited energy of sensor resources in real-world applicationallowing sensors to be shared between multiple tasks
to cope with the dynamic nature of WSNsconsidering the possibility of reassigning sensors
The Assignment Problem
In the network we have a set of sensors Each sensor is defined by its:
type, location and sensing range, the maximum utility it can provide, and the cost of using the sensor.
Missions may arrive at anytime and are collections of tasks.
Each task is defined by its: type, location and operational range, and demand, budget and profit
Each sensor-task assignment has an associated utility (the utility provided to the task by the sensor).
The Assignment Problem
Constraints on possible solutionsAll tasks within a mission must be satisfied for the
mission to be satisfiedThe utility achieved must greater than or equal to
the threshold for each task within a missionThe total cost of an assignment must be within
budgetThe set of sensor types of the sensors assigned to
must cover its information requirements Sensors cannot be assigned to more than one
type of task
Challenges
• A huge and dynamic number of constraints and variables
using SAM to reduce the search space
• The constraints form an instance of the Generalised Assignment Problem which is NP-Hard
our idea is to use a multi-round Knapsack-based algorithm since GAP can reduce to the Multiple Knapsack problem
• Finding solutions requires soft-real time; sensors are only partially observation about environment; the order of arrival of missions is unknown etc.
An agent-based approach is highly suited to the coordination of sensor resources in a decentralised and flexible manner
MSM
• MSM – Multiple Sensor Mode assignment mechanism
• Sensors are represented by agents• Sensor agents are cooperative• Each task is delegated to an agent
within the operational range• This agent acts as coordinator (not
necessarily involved in the solution)
MSM
• MSM operates as follows:– Coordinator identifies candidate sensors in
operational range and issues cfp– Each sensor makes independent decision
whether and what utility to bid– Coordinator attempts to allocate sensors
using GAP-E– If allocation fails, coord reports failure;
mission fails– Coord informs agents of allocations
GAP-E
• Each task has a priority ordering over sensor types (info requirements)
• Each task has a budget, allocated over sensor types• * Compute “cost matrix” for sensors on basis of bids
from sensors and priority over types• Run FPTAS algorithm• If no solution, seek sensor that can be released from
prior commitment to another task• If solution found within budget for all types, return• Recompute “cost matrix” and iterate from *
Experimental results
Hypothesis 1: MSM performs well in comparison to the estimated optimum
Mission success rate with 4 sensor types and 4missions arriving per hour
Mission success rate with 8 sensor types and 8missions arriving per hour
Experimental results
Hypothesis 2: The computational complexity (running time) of MSM is much less than that of other mechanisms
Running time (ms) with 4 sensor types and 4 missions arriving per hour
Running time (ms) with 8 sensor types and 8 missionsarriving per hour
Experimental results
Hypothesis 3: The computational complexity of MSM is increased in a steadily fashion with the number of missions (or tasks)
Running time (ms) with 4 sensor types and 25 sensorsper type
Future Work
Sensors are assumed to be static Tasks are independent Sensor agents are cooperative (will release a sensor
even if utility for its task is lower) We assume that tasks sharing a sensor require the
same information
Conclusion
A decentralised approach to solving the sensor-mission assignment problem for tasks sharing assets Generic solution to the resource allocation problem (both
sensors and tasks are heterogeneous) Sensor sharing significantly improves the number of
successfully allocated missions Use of polynomial algorithm within GAP-E increases
performance, and hence utility of solution in practical use Allows sensors to be reassigned to reduce effect of mission
arrival time on the solution