MASDIMA – Multi-Agent System for DIsruption MAnagementeol/SMA/20162017/ACastro... · NIAD&R –...
Transcript of MASDIMA – Multi-Agent System for DIsruption MAnagementeol/SMA/20162017/ACastro... · NIAD&R –...
NIAD&R – Distributed Artificial Intelligence and Robotics Group 1
Argumentation-Based Negotiation AN APPLICATION TO COOPERATIVE DISTRIBUTED PROBLEM SOLVING IN
OPERATIONAL CONTROL CENTRES
SMA/PRODEI (10-11-2011)
António Castro
NIAD&R – Distributed Artificial Intelligence and Robotics Group 2
• Motivation and Background: – Negotiation – Multi-agent Systems
• Operational Control Centres • GQ-Negotiation Protocol • Example of an ABN in AOCC • A PhD Dissertation Proposal
INDEX
NIAD&R – Distributed Artificial Intelligence and Robotics Group 3
Motivation and Background:
Negotiation
NIAD&R – Distributed Artificial Intelligence and Robotics Group 4
Negotiation is essential in settings where autonomous agents have conflicting interests
and a desire to cooperate
NEGOTIATION
Rahwan et al, 2004
NIAD&R – Distributed Artificial Intelligence and Robotics Group 5
A powerful paradigm to be used not only to reach agreements regarding more commons scenarios like buying and selling of goods but
also as a problem resolution technique
AUTOMATED NEGOTIATION
NIAD&R – Distributed Artificial Intelligence and Robotics Group 6
• Goal is to determine the optimal strategy by analysing the interaction as a game between identical participants and seeking its equilibrium
• Assumes participants behave according assumptions of rational-choice theory:
– Agents have unbounded computational resources
– Space of outcomes is completely known
AUTOMATED NEGOTIATION CLASSIFICATION
Game-theory Approaches
In most realistic environments we have bounded-rationaly, i.e., limited processing and communication capabilities as well as resource or time constraints.
NIAD&R – Distributed Artificial Intelligence and Robotics Group 7
• Various heuristic decision functions are used for evaluating and generating offers or proposals
• Relaxed assumptions about agent’s rationality and resources
• Fits better (most of) the realistic environments
• Agent’s utilities or preferences (usually) are assumed to be completely characterized apriori and fixed (like in previous approach).
AUTOMATED NEGOTIATION CLASSIFICATION
Heuristic-based Approaches
1. Usually leads to sub-optimal outcomes
2. Difficult to predict how the system and constituent agents will behave: – The models need extensive evaluation through simulations and empirical analysis
NIAD&R – Distributed Artificial Intelligence and Robotics Group 8
• Agents exchange additional information arguing about their beliefs during negotiation process
• In addition to accepting or rejecting a proposal:
– an agent can offer a critique, a justification,….
– Make a threat or promise a reward
– Change the negotiation object (new attributes or dimensions)
• Agent’s utilities and preferences might change during negotiation
• A more human-like negotiation approach
AUTOMATED NEGOTIATION CLASSIFICATION
Argumentation-based Approaches
These models are typically more complex than the game-theoretic and heuristic ones
NIAD&R – Distributed Artificial Intelligence and Robotics Group 9
Motivation and Background:
Agents, Multi-Agent Systems and Environments
NIAD&R – Distributed Artificial Intelligence and Robotics Group 10
• A powerful paradigm to represent an organization and their individuals.
• In LIACC/NIADR we use MAS and Negotiation in different domains, like:
– Partner selection in B2B E-commerce (Q-Negotiation).
– Problem resolution in an AOCC (GQ-Negotiation)
AGENTS AND MULTI-AGENT SYSTEMS
NIAD&R – Distributed Artificial Intelligence and Robotics Group 11
• Competitive Environment:
– Agents are selfish and compete with each other (e-commerce)
• Cooperative Environment:
– Agents do not have full knowledge/expertize to reach their goals, so, they need the cooperation of others.
• Distributed Environment:
– Physical
– Functional (different expertise)
– Spatial (local views)
ENVIRONMENTS
NIAD&R – Distributed Artificial Intelligence and Robotics Group 12
Operational Control Centres
NIAD&R – Distributed Artificial Intelligence and Robotics Group 13
AIR TRAFFIC CONTROL
Flight 103 enters the Porto airspace and declares that needs to land immediately due to lack of fuel. It has fuel for only 10 minutes.
At the same time there are 6 other aircrafts approaching the airport, each one with a specific level of fuel.
What is the best solution to solve this unexpected event?
NIAD&R – Distributed Artificial Intelligence and Robotics Group 14
AIRLINE OPERATIONS CONTROL
Flight 103 with a schedule ETA in Lisbon at 11:00 did not departure yet from Rio due to an aircraft malfunction. The new ETD is 20:00 and the new ETA is 12:30.
It has 230 passengers on board with the following connections:
20 to flight 231 – ETD 12:15
34 to flight 350 – ETD 12:00
60 to flight 412 – ETD 12:45
What is the best solution to solve this unexpected event?
NIAD&R – Distributed Artificial Intelligence and Robotics Group 15
INTEGRATED OPERATIONAL CONTROL CENTER
NIAD&R – Distributed Artificial Intelligence and Robotics Group 16
PROBLEM RESOLUTION PROCESS
NIAD&R – Distributed Artificial Intelligence and Robotics Group 17
AOCC EVENTS AND ACTIONS
Aircraft Malfunction => Flight Departure Delay • Swap the aircraft • Delay flight for the necessary time • Cancel flight and redirect passengers other flights • Rent an aircraft and crew from another company
Enroute Air Traffic Delay => Flight Arrival Delay • Use reserve crew to replace delayed one at dep. • Swap crew at departure. • Delay connection flights (if possible) • Redirect passengers with connections to others flights
NIAD&R – Distributed Artificial Intelligence and Robotics Group 18
Generic Q-Negotiation Protocol
NIAD&R – Distributed Artificial Intelligence and Robotics Group 19
• Multi-Attribute: – Each with a set of preferred values and domains
• Qualitative Feedback: – Organizer agent classifies the values of each attribute and gives
feedback
• Several Rounds • Inter-agents negotiation:
– Respondent agents negotiate in each round to be able to complete their knowledge and present a proposal
• Learning: – Agents adapt their strategy during proposal formulation
• Multiple agent types and roles: – Organizer, Respondent. Initiator, Participants.
• Suitable for competitive and cooperative environments
GQN CHARACTERISTICS
NIAD&R – Distributed Artificial Intelligence and Robotics Group 20
GQ-Negotiation
Event_time: 18-09-2009 12:39
Event_type: AC
Resource_Affected: CS-TTG
Resource_Type: A319
ETR: 110
Flt_date: 18-09-2009
Flt_Nbr: 686
STD: 13:20
STA: 15:50
ETA: 16:10
From: LIS
To: LUX
Dep_Delay: 20
Bus_Pax: 1
Econ_Pax: 128
Schd_Trip_Time: 02:30
Est_Trip_Time: 02:50
Schd_AC_Cost: 1084
Schd_Crew_Cost: 1745
Schd_Pax_Cost: 0
Send Problem
Prepares CFP according to preferences
CFP
Search Best Crew Solution
Search Best Pax Solution
Search Best AC Solution
Request pax and ac solution considering crew restrictions
Request crew and ac solution considering pax restrictions
Request crew and pax solution considering ac restrictions
Propose full solution
Propose
Propose
Evaluates bids and replies with Feedback according to preferences
Chooses winner
Ask Approval
ACTION, RESOURCE, RES_TYPE, DELAY, COST
NEAREST, 24164.6 , NB , 10 , 1900
ACTION, DELAY_TT, COST
KEEP , 20 , 2000
ACTION, RESOURCE, RES_TYPE, DELAY, COST
DELAY , CS-TTG , A319 , 20 , 1300
DLY_AC COST_AC DLY_CRW COST_CRW DLY_TT COST_PAX
20 980 10 1900 15 1500
DLY_AC COST_AC DLY_CRW COST_CRW DLY_TT COST_PAX
10 1230 0 1800 20 2000
DLY_AC COST_AC DLY_CRW COST_CRW DLY_TT COST_PAX
20 1300 10 1900 15 1500
Accepts
Close Problem
Does not Accept
Prepares new CFP considering human feedback (changing prefs)
Provides feedback
NIAD&R – Distributed Artificial Intelligence and Robotics Group 21
ABN – An Example in AOCC
NIAD&R – Distributed Artificial Intelligence and Robotics Group 22
NEGOTIATION
Argument
Interpretation
Argument
Generation
Argument
Selection
NIAD&R – Distributed Artificial Intelligence and Robotics Group 23
• Different types of arguments:
– An appeal to prevailing practice
– A counter example
– An appeal to past promise
– An appeal to self-interest
– A promise of future reward
– A threat
ARGUMENTS
weak
strong
NIAD&R – Distributed Artificial Intelligence and Robotics Group 24
ABN Protocol
Event_time: 18-09-2009 12:39
Event_type: AC
Resource_Affected: CS-TTG
Resource_Type: A319
ETR: 110
Flt_date: 18-09-2009
Flt_Nbr: 686
STD: 13:20
STA: 15:50
ETA: 16:10
From: LIS
To: LUX
Dep_Delay: 20
Bus_Pax: 1
Econ_Pax: 128
Schd_Trip_Time: 02:30
Est_Trip_Time: 02:50
Schd_AC_Cost: 1084
Schd_Crew_Cost: 1745
Schd_Pax_Cost: 0
Send Problem
Prepares CFP according to preferences
CFP
Search Best Crew Solution
Search Best Pax Solution
Search Best AC Solution
Propose full solution
Propose
Propose
Evaluates bids and replies using ABN
Chooses winner
Close Problem
1) PROPOSE: KEEP,20,2000,(APPEAL_SELFI: VIP PAX)
2) INFORM: (APPEAL_CURRPRT: MIN COSTS)
Inform
3) INFORM: (THREAT: VIP WILL NEVER FLY IN THIS COMPANY)
Inform
4) INFORM: (ARG_ACCEPTED)
Inform
NIAD&R – Distributed Artificial Intelligence and Robotics Group 25
A PhD Dissertation Proposal
NIAD&R – Distributed Artificial Intelligence and Robotics Group 26
PRODEI – PHD DISSERTATION PROPOSAL
Cooperative Distributed Problem Solving in Operational Control Centres AN ARGUMENTATION-BASED NEGOTIATION APPROACH
Supervisor
Prof. Ana Paula Rocha (LIACC/FEUP)
Co-supervisor
António Castro (LIACC/TAP Portugal)
NIAD&R – Distributed Artificial Intelligence and Robotics Group 27
The definition of a methodology and negotiation protocol that includes the ABN components as a way of solving problems in operational control
centres
THE RESEARCH PROPOSAL IN ONE PARAGRAPH
NIAD&R – Distributed Artificial Intelligence and Robotics Group 28
– Argument Interpretation • Dealing with arguments from multiple sources
– Argument Generation (how?) • Dealing with multiple issues
• Learning with the past (e.g, CBR)
• Using context information
• Using profile information
– Argument Selection (how?) • Based on agent utility
• Based on argument strength
• Based on the counterparts’ reply
RESEARCH QUESTIONS….
NIAD&R – Distributed Artificial Intelligence and Robotics Group 29
The research necessary for this PhD proposal has the cooperation of TAP Portugal
NIAD&R – Distributed Artificial Intelligence and Robotics Group 30
• Rahwan, et al., (2004). Argumentation-Based Negotiation. The Knowledge Engineering Review. Vol. 18:4, pp. 343-375. Cambridge University Press.
• Rocha, A.P. and Oliveira, E. (2001). Electronic Institutions as a framework for Agents' Negotiation and mutual Commitment, in Progress in Artificial Intelligence (Proceedings of 10th EPIA) , Eds. P.Brazdil, A.Jorge, LNAI 2258, pp.232-245,Springer.
• Castro, A.J.M. and Oliveira, E. (2011). A New Concept for Disruption Management in Airline Operations Control. Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering, Vol. 225, N. 3 (March 2011), pp. 269-290.
• Jin, Y. and Geslin, M. (2009). Argumentation-based negotiation for collaborative engineering design. Int. Journal Collaborative Engineering, Vol. 1, N. ½, pp. 125-151.
BIBLIOGRAPHIC REFERENCES
NIAD&R – Distributed Artificial Intelligence and Robotics Group 31
• Thank you very much!
• Questions?
THE END
Links e E-mails
MASDIMA Project: http://www.disruptionmanagement.com Ana Paula Rocha [email protected]
António Castro [email protected]