Rubrik

1
CORA Advanced Search for Decision Support
Anders Törne
2

• CORA - Assisting Decisions
• an example from Air Traffic Management
• The Conflict Resolution Assistant CORA2

Representing the decision? What is a good
decision? Finding proposal(s) for
decisions? Presenting the proposal(s)?
3
Conflict Resolution AssistantCORA
Conflict gt two aircrafts at same flight level.
Planned 4D trajectory positions separated less
than 5 nautical miles.
Pilot FMS System
Other Sectors - Air Traffic Controllers
ATM System
A resolution solves the conflict
20 min lead. Conflict detection by trajectory
planning.
Air Traffic Controller
4
CORA - expressing the decision example with
three manoeuvres
5
CORA - representation of resolutions the
manoeuvre alphabet

• The resolution from the earlier example isthe
  manoeuvre sequence ( (TL, A/C1, t1, 15 ), (SI,
  A/C2, t2, M.02), (DR, A/C1, t3, XCOR4) )
• The manoeuvre alphabet - each place/variable in
  the manoeuvre representation may take a discrete
  set of values. These values spans the manoeuvre
  domain.
• The language for resolutions primitives
  (manoeuvres) and means of combination
  (sequencing), but means of abstractions not
  required

6
CORA - what is a good resolution? 1 (2)

• A resolution must be valid it must solve the
  conflict and be possible to manoeuvre for the
  A/C
• Finding a good, valid resolution is a
  multi-objective optimization problem a
  calculatable cost must be found for each
  objective and a good resolution is one with a low
  cost
• Too many objectives makes the optimization
  process unstable and a divide and conquer
  strategy is required

7
CORA - what is a good resolution? 2 (2)

• The following fairly independent objectives are
  used to calculate cost in the top optimization
  process.
• IncreaseSafety
• DecreaseControllerLoad
• DecreasePilotLoad
• IncreaseAirlineSatisfaction

8
Partitioning the objectives into subobjectives 1
(2)

• IncreaseSafety consist of the subobjectives
• AvoidConflict
• AirspaceRules
• DecreaseControllerLoad consist of the
  subobjectives
• MoreThanOne
• DecisionLoad
• MonitoringLoad
• SequenceLoad
• ClearanceLoad
• CoordinationLoad

9
Partitioning the objectives into subobjectives 2
(2)

• DecreasePilotLoad consist of the subobjectives
• PilotTime
• PilotLogic
• AvoidDisturbance
• IncreaseAirlineSatisfaction consist of the
  subobjectives
• ImproveEconomy
• ImproveService

10
Finding proposal(s) for resolutions

• Discrete search algorithms large field
• Uninformed search (e.g., brute force tree
  search),
• Informed search uses heuristics A,
• Constraint satisfaction, Genetic algorithms,
  Simulated annealing
• Tabu-search
• Simple variant chosen A

11
A - search tree
12
Algorithm queues sorted according to cost
Expand node
Result_container
Pop_first_node
Sorted_node_queue
Filter away non-valid
Left Turn
Sort_insert
For each remaining node Node calculations
Direct
For each remaining node - CostCalc
Speed Inc
. . .
For each remaining node - Complete?
No
Search_queue
Etc.
Yes
Save_node - it is a proposal
Queueofqueues
13
Trajectory/Resolution example
Resolution manoeuvre
FirstRefPoint
Original manoeuvre
SpeedIncrease
Tail
Resolution Trajectory
LeftTurn
LevelIncrease with time assigned
SecondRefPoint
14
FLOS First Loss of SeparationELOS End of
Loss of SeparationCPA Closest point of approach
15
CORA2 CONTEXT
DMAN Departure Manager
MTCD - Medium Term Conflict Detector
SUPERVISOR
AMAN-Arrival Manager
EMAN En-route Manager
FDPS Flight Data Processing System
Env DB Environment Database
SYSCO System Co-ordinator
CLOCK
TED Trajectory Editor
CA Clearance Assistant
16
Other domains similar searchBest-route search
3D model Lasersystem, FOI3D vizualization
MSI, FOICalculation time - minFind cheapest
route
Multi objective search
17
CORA - summary

• Decision support
• Discrete search for feasible solution in large
  solution space
• Example CORA (Path finding)
• Multi-objective optimization
• FOI offers expertise in optimization and planning
  for decision support